diff --git a/articles/aorsf.html b/articles/aorsf.html index 622d6c7d..8fd9bf3e 100644 --- a/articles/aorsf.html +++ b/articles/aorsf.html @@ -137,7 +137,7 @@

Oblique RFs for #> N trees: 500 #> N predictors total: 7 #> N predictors per node: 3 -#> Average leaves per tree: 5.648 +#> Average leaves per tree: 5.544 #> Min observations in leaf: 5 #> OOB stat value: 1.00 #> OOB stat type: AUC-ROC @@ -158,9 +158,9 @@

Oblique RFs for #> N trees: 500 #> N predictors total: 7 #> N predictors per node: 3 -#> Average leaves per tree: 49.674 +#> Average leaves per tree: 49.872 #> Min observations in leaf: 5 -#> OOB stat value: 0.82 +#> OOB stat value: 0.81 #> OOB stat type: RSQ #> Variable importance: anova #> @@ -185,7 +185,7 @@

Oblique RFs for #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 -#> Average leaves per tree: 22 +#> Average leaves per tree: 21.4 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.77 @@ -226,14 +226,14 @@

Variable importance
 
 orsf_vi_negate(pbc_fit)
-#>        copper         stage       protime           sex       ascites 
-#>  0.1188219833  0.0665535400  0.0464884356  0.0384596473  0.0344990821 
-#>        hepato          bili           ast          trig       albumin 
-#>  0.0317538618  0.0277185671  0.0271787128  0.0246554066  0.0217429643 
-#>           age          chol      alk.phos           trt       spiders 
-#>  0.0150528840  0.0126460875  0.0026088161  0.0016801827  0.0007598719 
-#>         edema      platelet 
-#> -0.0006134379 -0.0066584545
+#> bili spiders copper albumin sex +#> 0.1671647745 0.0520379071 0.0432409174 0.0292245433 0.0258174901 +#> alk.phos trt protime ast stage +#> 0.0207262097 0.0196654076 0.0147894928 0.0138953587 0.0094377622 +#> age ascites trig edema platelet +#> 0.0063243985 0.0053495205 0.0021405459 0.0018460367 0.0008203822 +#> hepato chol +#> 0.0004910894 -0.0128635357

  • You can also compute variable importance using @@ -243,10 +243,10 @@

    Variable importance
     
 orsf_vi_permute(penguin_fit)
-#>    bill_length_mm flipper_length_mm     bill_depth_mm       body_mass_g 
-#>       0.169789864       0.100880436       0.076946210       0.067635877 
-#>            island               sex              year 
-#>       0.061011929       0.019879728       0.001228741
    +#> bill_length_mm flipper_length_mm bill_depth_mm island +#> 0.1738256958 0.0898851835 0.0810334484 0.0655328821 +#> body_mass_g sex year +#> 0.0647936728 0.0191948220 0.0005500798

  • A faster alternative to permutation and negation importance is @@ -256,9 +256,9 @@

    Variable importance orsf_vi_anova(bill_fit) #> species sex island flipper_length_mm -#> 0.34412460 0.21024699 0.11541213 0.08715245 +#> 0.35859861 0.20116912 0.11896429 0.08734940 #> body_mass_g bill_depth_mm year -#> 0.07891657 0.06775045 0.01330396 +#> 0.08107428 0.06459390 0.01368270

    • diff --git a/articles/fast.html b/articles/fast.html index a8ecc848..70697e33 100644 --- a/articles/fast.html +++ b/articles/fast.html @@ -126,8 +126,8 @@

    Don’t specify a control# unspecified control is much faster time_net['elapsed'] / time_fast['elapsed'] -#> elapsed -#> 45.43478 +#> elapsed +#> 44.125

    Use n_thread @@ -218,11 +218,11 @@

    Don’t wait. Estimate! time_est <- orsf_time_to_train(fit_spec, n_tree_subset = 5) ) #> user system elapsed -#> 0.268 0.008 0.275 +#> 0.286 0.020 0.307 # the estimated training time: time_est -#> Time difference of 110.1083 secs

    +#> Time difference of 122.3736 secs diff --git a/articles/oobag.html b/articles/oobag.html index fd1f5cda..23a78463 100644 --- a/articles/oobag.html +++ b/articles/oobag.html @@ -139,9 +139,9 @@

    Out-of-bag predictions and error# what is the output from this function? fit$eval_oobag$stat_values #> [,1] -#> [1,] 0.7737671 +#> [1,] 0.7652389

    The out-of-bag estimate of Harrell’s C-index (the default method to -evaluate out-of-bag predictions) is 0.7737671.

    +evaluate out-of-bag predictions) is 0.7652389.

    Monitoring out-of-bag error diff --git a/articles/oobag_files/figure-html/unnamed-chunk-2-1.png b/articles/oobag_files/figure-html/unnamed-chunk-2-1.png index f9ac3c29..f736cad9 100644 Binary files a/articles/oobag_files/figure-html/unnamed-chunk-2-1.png and b/articles/oobag_files/figure-html/unnamed-chunk-2-1.png differ diff --git a/pkgdown.yml b/pkgdown.yml index 04a37c2d..98dce1de 100644 --- a/pkgdown.yml +++ b/pkgdown.yml @@ -6,7 +6,7 @@ articles: fast: fast.html oobag: oobag.html pd: pd.html -last_built: 2024-01-15T18:17Z +last_built: 2024-01-15T23:36Z urls: reference: https://bcjaeger.github.io/aorsf/reference article: https://bcjaeger.github.io/aorsf/articles diff --git a/reference/orsf_summarize_uni.html b/reference/orsf_summarize_uni.html index a070209f..44afd445 100644 --- a/reference/orsf_summarize_uni.html +++ b/reference/orsf_summarize_uni.html @@ -157,20 +157,22 @@

    Examples orsf_summarize_uni(object, n_variables = 2) #> -#> -- ascites (VI Rank: 1) ------------------------ +#> -- bili (VI Rank: 1) --------------------------- #> #> |---------------- Risk ----------------| -#> Value Mean Median 25th % 75th % -#> 0 0.3112269 0.2184047 0.04443568 0.535574 -#> 1 0.4763575 0.3918262 0.24704845 0.714196 +#> Value Mean Median 25th % 75th % +#> 0.60 0.2313780 0.1479413 0.04781748 0.3280577 +#> 0.80 0.2364004 0.1493642 0.05583256 0.3366922 +#> 1.40 0.2650110 0.2064121 0.07340697 0.3652209 +#> 3.52 0.3713478 0.3347867 0.17337079 0.5443745 +#> 7.25 0.4685566 0.4452742 0.28893537 0.6351033 #> -#> -- edema (VI Rank: 2) -------------------------- +#> -- ascites (VI Rank: 2) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % -#> 0 0.3054237 0.1884637 0.04443568 0.5342492 -#> 0.5 0.3531583 0.2543017 0.05655187 0.5969800 -#> 1 0.4559456 0.4038856 0.23603721 0.6813184 +#> 0 0.2999651 0.2009324 0.06215675 0.5083512 +#> 1 0.5279761 0.5040944 0.37091543 0.7086838 #> #> Predicted risk at time t = 1788 for top 2 predictors @@ -184,21 +186,18 @@

    Examples#> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % -#> 0.60 0.2224069 0.1254764 0.03301986 0.3207543 -#> 0.80 0.2262946 0.1254764 0.03266947 0.3316953 -#> 1.40 0.2536990 0.1646659 0.05763091 0.3902082 -#> 3.52 0.3911796 0.3181171 0.17745485 0.6078190 -#> 7.25 0.5031346 0.4685324 0.33257259 0.6883006 +#> 0.60 0.2313780 0.1479413 0.04781748 0.3280577 +#> 0.80 0.2364004 0.1493642 0.05583256 0.3366922 +#> 1.40 0.2650110 0.2064121 0.07340697 0.3652209 +#> 3.52 0.3713478 0.3347867 0.17337079 0.5443745 +#> 7.25 0.4685566 0.4452742 0.28893537 0.6351033 #> -#> -- copper (VI Rank: 2) ------------------------- +#> -- sex (VI Rank: 2) ---------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % -#> 25.5 0.2697228 0.1414728 0.03983757 0.4498140 -#> 42.8 0.2709936 0.1391208 0.03839266 0.4498140 -#> 74.0 0.2865622 0.1740328 0.04594795 0.4755196 -#> 129 0.3404841 0.2498839 0.09485429 0.5258744 -#> 214 0.4435109 0.3924181 0.20542624 0.6589764 +#> m 0.3319983 0.2533052 0.11251192 0.5482338 +#> f 0.3037761 0.1968855 0.06036558 0.5263657 #> #> Predicted risk at time t = 1788 for top 2 predictors diff --git a/reference/orsf_time_to_train.html b/reference/orsf_time_to_train.html index 51c4a01e..f00bb294 100644 --- a/reference/orsf_time_to_train.html +++ b/reference/orsf_time_to_train.html @@ -96,7 +96,7 @@

    Examplestime_estimated <- orsf_time_to_train(object, n_tree_subset=1) print(time_estimated) -#> Time difference of 0.04493713 secs +#> Time difference of 0.04523039 secs # let's see how close the approximation was time_true_start <- Sys.time() @@ -106,11 +106,11 @@

    Examplestime_true <- time_true_stop - time_true_start print(time_true) -#> Time difference of 0.02969646 secs +#> Time difference of 0.0303793 secs # error abs(time_true - time_estimated) -#> Time difference of 0.01524067 secs +#> Time difference of 0.01485109 secs

    diff --git a/search.json b/search.json index 7d76dca0..d2cbab5e 100644 --- a/search.json +++ b/search.json @@ -1 +1 @@ -[{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":null,"dir":"","previous_headings":"","what":"Contributing to aorsf","title":"Contributing to aorsf","text":"Want contribute aorsf? Great! aorsf initially stable state development, great deal active subsequent development envisioned. outline propose change aorsf. detailed info contributing , tidyverse packages, please see development contributing guide.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"fixing-typos","dir":"","previous_headings":"","what":"Fixing typos","title":"Contributing to aorsf","text":"can fix typos, spelling mistakes, grammatical errors documentation directly using GitHub web interface, long changes made source file. generally means ’ll need edit roxygen2 comments .R, .Rd file. can find .R file generates .Rd reading comment first line.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"bigger-changes","dir":"","previous_headings":"","what":"Bigger changes","title":"Contributing to aorsf","text":"want make bigger change, ’s good idea first file issue make sure someone team agrees ’s needed. ’ve found bug, please file issue illustrates bug minimal reprex (also help write unit test, needed).","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"pull-request-process","dir":"","previous_headings":"Bigger changes","what":"Pull request process","title":"Contributing to aorsf","text":"Fork package clone onto computer. haven’t done , recommend using usethis::create_from_github(\"ropensci/aorsf\", fork = TRUE). Install development dependencies devtools::install_dev_deps(), make sure package passes R CMD check running devtools::check(). R CMD check doesn’t pass cleanly, ’s good idea ask help continuing. Create Git branch pull request (PR). recommend using usethis::pr_init(\"brief-description--change\"). Make changes, commit git, create PR running usethis::pr_push(), following prompts browser. title PR briefly describe change. body PR contain Fixes #issue-number. user-facing changes, add bullet top NEWS.md (.e. just first header). Follow style described https://style.tidyverse.org/news.html.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"code-style","dir":"","previous_headings":"Bigger changes","what":"Code style","title":"Contributing to aorsf","text":"New code follow tidyverse style guide. can use styler package apply styles, please don’t restyle code nothing PR. use roxygen2, Markdown syntax, documentation. use testthat unit tests. Contributions test cases included easier accept.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"code-of-conduct","dir":"","previous_headings":"","what":"Code of Conduct","title":"Contributing to aorsf","text":"Please note aorsf project released Contributor Code Conduct. contributing project agree abide terms.","code":""},{"path":"https://bcjaeger.github.io/aorsf/LICENSE.html","id":null,"dir":"","previous_headings":"","what":"MIT License","title":"MIT License","text":"Copyright (c) 2022 aorsf authors (Byron C. Jaeger, Sawyer Welden, Nicholas M. Pajewski) Permission hereby granted, free charge, person obtaining copy software associated documentation files (“Software”), deal Software without restriction, including without limitation rights use, copy, modify, merge, publish, distribute, sublicense, /sell copies Software, permit persons Software furnished , subject following conditions: copyright notice permission notice shall included copies substantial portions Software. SOFTWARE PROVIDED “”, WITHOUT WARRANTY KIND, EXPRESS IMPLIED, INCLUDING LIMITED WARRANTIES MERCHANTABILITY, FITNESS PARTICULAR PURPOSE NONINFRINGEMENT. EVENT SHALL AUTHORS COPYRIGHT HOLDERS LIABLE CLAIM, DAMAGES LIABILITY, WHETHER ACTION CONTRACT, TORT OTHERWISE, ARISING , CONNECTION SOFTWARE USE DEALINGS SOFTWARE.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"background","dir":"Articles","previous_headings":"","what":"Background","title":"Introduction to aorsf","text":"oblique random forest (RF) extension traditional (axis-based) RF. Instead using single variable split data grow new branches, trees oblique RF use weighted combination multiple variables.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"oblique-rfs-for-survival-classification-and-regression","dir":"Articles","previous_headings":"","what":"Oblique RFs for survival, classification, and regression","title":"Introduction to aorsf","text":"purpose aorsf (‘’ short accelerated) provide unifying framework fit oblique RFs can scale adequately large data sets. fastest algorithms available package used default often equivalent prediction accuracy computational approaches. center piece aorsf orsf() function. initial versions aorsf, orsf() function fit oblique random survival forests, now allows classification, regression, survival forests. (may introduce orf() function future name orsf() misleading users.) classification, fit oblique RF predict penguin species using penguin data magnificent palmerpenguins R package regression, use data predict bill length penguins: personal favorite oblique survival RF accelerated Cox regression great combination prediction accuracy computational efficiency (see JCGS paper). , predict mortality risk following diagnosis primary biliary cirrhosis: may notice first input aorsf data. design choice makes easier use orsf pipes (.e., %>% |>). instance,","code":"# An oblique classification RF penguin_fit <- orsf(data = penguins_orsf, formula = species ~ .) penguin_fit #> ---------- Oblique random classification forest #> #> Linear combinations: Accelerated Logistic regression #> N observations: 333 #> N classes: 3 #> N trees: 500 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 5.648 #> Min observations in leaf: 5 #> OOB stat value: 1.00 #> OOB stat type: AUC-ROC #> Variable importance: anova #> #> ----------------------------------------- # An oblique regression RF bill_fit <- orsf(data = penguins_orsf, formula = bill_length_mm ~ .) bill_fit #> ---------- Oblique random regression forest #> #> Linear combinations: Accelerated Linear regression #> N observations: 333 #> N trees: 500 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 49.674 #> Min observations in leaf: 5 #> OOB stat value: 0.82 #> OOB stat type: RSQ #> Variable importance: anova #> #> ----------------------------------------- # An oblique survival RF pbc_fit <- orsf(data = pbc_orsf, n_tree = 5, formula = Surv(time, status) ~ . - id) pbc_fit #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 22 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.77 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> ----------------------------------------- library(dplyr) pbc_fit <- pbc_orsf |> select(-id) |> orsf(formula = Surv(time, status) ~ ., n_tree = 5)"},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"interpretation","dir":"Articles","previous_headings":"","what":"Interpretation","title":"Introduction to aorsf","text":"aorsf includes several functions dedicated interpretation ORSFs, estimation partial dependence variable importance.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"variable-importance","dir":"Articles","previous_headings":"Interpretation","what":"Variable importance","title":"Introduction to aorsf","text":"multiple methods compute variable importance, can applied type oblique forest. compute negation importance, ORSF multiplies coefficient variable -1 re-computes --sample (sometimes referred --bag) accuracy ORSF model. can also compute variable importance using permutation, classical approach noises predictor assigned resulting degradation prediction accuracy importance predictor. faster alternative permutation negation importance ANOVA importance, computes proportion times variable obtains low p-value (p < 0.01) forest grown.","code":"orsf_vi_negate(pbc_fit) #> copper stage protime sex ascites #> 0.1188219833 0.0665535400 0.0464884356 0.0384596473 0.0344990821 #> hepato bili ast trig albumin #> 0.0317538618 0.0277185671 0.0271787128 0.0246554066 0.0217429643 #> age chol alk.phos trt spiders #> 0.0150528840 0.0126460875 0.0026088161 0.0016801827 0.0007598719 #> edema platelet #> -0.0006134379 -0.0066584545 orsf_vi_permute(penguin_fit) #> bill_length_mm flipper_length_mm bill_depth_mm body_mass_g #> 0.169789864 0.100880436 0.076946210 0.067635877 #> island sex year #> 0.061011929 0.019879728 0.001228741 orsf_vi_anova(bill_fit) #> species sex island flipper_length_mm #> 0.34412460 0.21024699 0.11541213 0.08715245 #> body_mass_g bill_depth_mm year #> 0.07891657 0.06775045 0.01330396"},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"partial-dependence-pd","dir":"Articles","previous_headings":"Interpretation","what":"Partial dependence (PD)","title":"Introduction to aorsf","text":"Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model’s prediction. PD, see vignette","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"individual-conditional-expectations-ice","dir":"Articles","previous_headings":"Interpretation","what":"Individual conditional expectations (ICE)","title":"Introduction to aorsf","text":"Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor. ICE, see vignette","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"what-about-the-original-orsf","dir":"Articles","previous_headings":"","what":"What about the original ORSF?","title":"Introduction to aorsf","text":"original ORSF (.e., obliqueRSF) used glmnet find linear combinations inputs. aorsf allows users implement approach using orsf_control_survival(method = 'net') function: net forests fit lot faster original ORSF function obliqueRSF. However, net forests still much slower cph ones.","code":"orsf_net <- orsf(data = pbc_orsf, formula = Surv(time, status) ~ . - id, control = orsf_control_survival(method = 'net'))"},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"aorsf-and-other-machine-learning-software","dir":"Articles","previous_headings":"","what":"aorsf and other machine learning software","title":"Introduction to aorsf","text":"unique feature aorsf fast algorithms fit ORSF ensembles. RLT obliqueRSF fit oblique random survival forests, aorsf faster. ranger randomForestSRC fit survival forests, neither package supports oblique splitting. obliqueRF fits oblique random forests classification regression, survival. PPforest fits oblique random forests classification survival. Note: default prediction behavior aorsf models produce predicted risk specific prediction horizon, default ranger randomForestSRC. think change future, computing time independent predictions aorsf helpful.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"learning-more","dir":"Articles","previous_headings":"","what":"Learning more","title":"Introduction to aorsf","text":"aorsf began dedicated package oblique random survival forests, papers published far focused survival analysis risk prediction. However, routines regression classification oblique RFs aorsf high overlap survival ones. See orsf details oblique random survival forests. see JCGS paper details algorithms used specifically aorsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"go-faster","dir":"Articles","previous_headings":"","what":"Go faster","title":"Tips to speed up computation","text":"Analyses can slow crawl models need hours run. article find tricks prevent bottleneck using orsf().","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"dont-specify-a-control","dir":"Articles","previous_headings":"","what":"Don’t specify a control","title":"Tips to speed up computation","text":"default control orsf() NULL , unspecified, orsf() pick fastest possible control depending type forest grown. default control run-time compared approaches can striking. example:","code":"time_fast <- system.time( expr = orsf(pbc_orsf, formula = time+status~. -id, n_tree = 5) ) time_net <- system.time( expr = orsf(pbc_orsf, formula = time+status~. -id, control = orsf_control_survival(method = 'net'), n_tree = 5) ) # unspecified control is much faster time_net['elapsed'] / time_fast['elapsed'] #> elapsed #> 45.43478"},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"use-n_thread","dir":"Articles","previous_headings":"","what":"Use n_thread","title":"Tips to speed up computation","text":"n_thread argument uses multi-threading run aorsf functions parallel possible. know many threads want, e.g. want exactly 5, set n_thread = 5. aren’t sure many threads available want use feasible amount, using n_thread = 0 (default) tells aorsf . Note: sometimes multi-threading possible. example, R single threaded language, multi-threading applied orsf() needs call R functions C++, occurs customized R function used find linear combination variables compute prediction accuracy.","code":"# automatically pick number of threads based on amount available orsf(pbc_orsf, formula = time+status~. -id, n_tree = 5, n_thread = 0)"},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"do-less","dir":"Articles","previous_headings":"","what":"Do less","title":"Tips to speed up computation","text":"inputs orsf() can adjusted make run faster: set n_retry 0 set oobag_pred_type 'none' set importance 'none' increase split_min_events, split_min_obs, leaf_min_events, leaf_min_obs make trees stop growing sooner increase split_min_stat enforce strict requirements growing deeper trees. Applying tips: modifying inputs can make orsf() run faster, can also impact prediction accuracy.","code":"orsf(pbc_orsf, formula = time+status~., n_thread = 0, n_tree = 5, n_retry = 0, oobag_pred_type = 'none', importance = 'none', split_min_events = 20, leaf_min_events = 10, split_min_stat = 10)"},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"show-progress","dir":"Articles","previous_headings":"","what":"Show progress","title":"Tips to speed up computation","text":"Setting verbose_progress = TRUE doesn’t make anything run faster, can help make feel like things running less slow.","code":"verbose_fit <- orsf(pbc_orsf, formula = time+status~. -id, n_tree = 5, verbose_progress = TRUE) #> Growing trees: 100%. #> Computing predictions: 100%."},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"dont-wait--estimate","dir":"Articles","previous_headings":"","what":"Don’t wait. Estimate!","title":"Tips to speed up computation","text":"Instead running model hoping fast, can estimate long specification model take using no_fit = TRUE call orsf().","code":"fit_spec <- orsf(pbc_orsf, formula = time+status~. -id, control = orsf_control_survival(method = 'net'), n_tree = 2000, no_fit = TRUE) # how much time it takes to estimate training time: system.time( time_est <- orsf_time_to_train(fit_spec, n_tree_subset = 5) ) #> user system elapsed #> 0.268 0.008 0.275 # the estimated training time: time_est #> Time difference of 110.1083 secs"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"out-of-bag-data","dir":"Articles","previous_headings":"","what":"Out-of-bag data","title":"Out-of-bag predictions and evaluation","text":"random forests, tree grown bootstrapped version training set. bootstrap samples selected replacement, bootstrapped training set contains two-thirds instances original training set. ‘--bag’ data instances bootstrapped training set.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"out-of-bag-predictions-and-error","dir":"Articles","previous_headings":"","what":"Out-of-bag predictions and error","title":"Out-of-bag predictions and evaluation","text":"tree random forest can make predictions --bag data, --bag predictions can aggregated make ensemble --bag prediction. Since --bag data used grow tree, accuracy ensemble --bag predictions approximate generalization error random forest. --bag prediction error plays central role routines estimate variable importance, e.g. negation importance. fit oblique random survival forest plot distribution ensemble --bag predictions. Next, let’s check --bag accuracy fit: --bag estimate Harrell’s C-index (default method evaluate --bag predictions) 0.7737671.","code":"fit <- orsf(data = pbc_orsf, formula = Surv(time, status) ~ . - id, oobag_pred_type = 'surv', n_tree = 5, oobag_pred_horizon = 2000) hist(fit$pred_oobag, main = 'Out-of-bag survival predictions at t=2,000') # what function is used to evaluate out-of-bag predictions? fit$eval_oobag$stat_type #> [1] \"Harrell's C-index\" # what is the output from this function? fit$eval_oobag$stat_values #> [,1] #> [1,] 0.7737671"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"monitoring-out-of-bag-error","dir":"Articles","previous_headings":"","what":"Monitoring out-of-bag error","title":"Out-of-bag predictions and evaluation","text":"--bag data set contains one-third training set, --bag error estimate usually converges stable value trees added forest. want monitor convergence --bag error oblique random survival forest, can set oobag_eval_every compute --bag error every oobag_eval_every tree. example, let’s compute --bag error fitting tree forest 50 trees: general, least 500 trees recommended random forest fit. ’re just using 10 illustration.","code":"fit <- orsf(data = pbc_orsf, formula = Surv(time, status) ~ . - id, n_tree = 20, tree_seeds = 2, oobag_pred_type = 'surv', oobag_pred_horizon = 2000, oobag_eval_every = 1) plot( x = seq(1, 20, by = 1), y = fit$eval_oobag$stat_values, main = 'Out-of-bag C-statistic computed after each new tree is grown.', xlab = 'Number of trees grown', ylab = fit$eval_oobag$stat_type ) lines(x=seq(1, 20), y = fit$eval_oobag$stat_values)"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"user-supplied-out-of-bag-evaluation-functions","dir":"Articles","previous_headings":"","what":"User-supplied out-of-bag evaluation functions","title":"Out-of-bag predictions and evaluation","text":"cases, may want control --bag error estimated. example, let’s use Brier score SurvMetrics package: two ways apply function compute --bag error. First, can apply function --bag survival predictions stored ‘aorsf’ objects, e.g: Second, can pass function orsf(), used place Harrell’s C-statistic:","code":"oobag_brier_surv <- function(y_mat, w_vec, s_vec){ # output is numeric vector of length 1 as.numeric( SurvMetrics::Brier( object = Surv(time = y_mat[, 1], event = y_mat[, 2]), pre_sp = s_vec, # t_star in Brier() should match oob_pred_horizon in orsf() t_star = 2000 ) ) } oobag_brier_surv(y_mat = pbc_orsf[,c('time', 'status')], s_vec = fit$pred_oobag) #> [1] 0.11869 # instead of copy/pasting the modeling code and then modifying it, # you can just use orsf_update. fit_brier <- orsf_update(fit, oobag_fun = oobag_brier_surv) plot( x = seq(1, 20, by = 1), y = fit_brier$eval_oobag$stat_values, main = 'Out-of-bag error computed after each new tree is grown.', sub = 'For the Brier score, lower values indicate more accurate predictions', xlab = 'Number of trees grown', ylab = \"Brier score\" ) lines(x=seq(1, 20), y = fit_brier$eval_oobag$stat_values)"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"specific-instructions-on-user-supplied-functions","dir":"Articles","previous_headings":"User-supplied out-of-bag evaluation functions","what":"Specific instructions on user-supplied functions","title":"Out-of-bag predictions and evaluation","text":"use oobag_fun note following: oobag_fun three inputs: y_mat, w_vec, s_vec survival trees, y_mat two column matrix first column named ‘time’ second named ‘status’. classification trees, y_mat matrix number columns = number distinct classes outcome. regression, y_mat matrix one column. s_vec numeric vector containing predictions oobag_fun return numeric output length 1","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"notes","dir":"Articles","previous_headings":"","what":"Notes","title":"Out-of-bag predictions and evaluation","text":"evaluating --bag error: oobag_pred_horizon input orsf() determines prediction horizon --bag predictions. prediction horizon needs specified evaluate prediction accuracy cases, examples . sure check case using functions, , sure oobag_pred_horizon matches prediction horizon used custom function. functions expect predicted risk (.e., 1 - predicted survival), others expect predicted survival.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"partial-dependence-pd","dir":"Articles","previous_headings":"","what":"Partial dependence (PD)","title":"PD and ICE curves with ORSF","text":"Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model’s prediction. can compute PD individual conditional expectation (ICE) three ways: using -bag predictions training data. -bag PD indicates relationships model learned training. helpful goal interpret model. using --bag predictions training data. --bag PD indicates relationships model learned training using --bag data simulates application model new data. helpful want test model’s reliability fairness new data don’t access large testing set. using predictions new set data. New data PD shows model predicts outcomes observations seen. helpful want test model’s reliability fairness.","code":"library(aorsf) library(ggplot2)"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"classification","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Classification","title":"PD and ICE curves with ORSF","text":"Begin fitting oblique classification random forest: Compute PD using --bag data flipper_length_mm = c(190, 210). Note predicted probabilities returned class probabilities mean column sum 1 take sum class specific value pred_spec variables. example, isn’t case median predicted probability!","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_oob <- orsf_pd_oob(fit_clsf, pred_spec = pred_spec) pd_oob #> class flipper_length_mm mean lwr medn upr #> 1: Adelie 190 0.6182417 0.206899034 0.75537171 0.9796439 #> 2: Adelie 210 0.4348386 0.019519733 0.56802082 0.8620694 #> 3: Chinstrap 190 0.2114905 0.018420139 0.15561560 0.7174734 #> 4: Chinstrap 210 0.1806274 0.020409141 0.09928047 0.6990198 #> 5: Gentoo 190 0.1702678 0.001281382 0.02830728 0.5733438 #> 6: Gentoo 210 0.3845340 0.072260715 0.20258335 0.9519486 sum(pd_oob[flipper_length_mm == 190, mean]) #> [1] 1 sum(pd_oob[flipper_length_mm == 190, medn]) #> [1] 0.9392946"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"regression","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Regression","title":"PD and ICE curves with ORSF","text":"Begin fitting oblique regression random forest: Compute PD using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new #> flipper_length_mm mean lwr medn upr #> 1: 190 42.96571 37.09805 43.69769 48.72301 #> 2: 210 45.66012 40.50693 46.31577 51.65163 pred_spec = pred_spec_auto(species, island, body_mass_g) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new #> species island body_mass_g mean lwr medn upr #> 1: Adelie Biscoe 3200 40.31374 37.24373 40.31967 44.22824 #> 2: Chinstrap Biscoe 3200 45.10582 42.63342 45.10859 47.60119 #> 3: Gentoo Biscoe 3200 42.81649 40.19221 42.55664 46.84035 #> 4: Adelie Dream 3200 40.16219 36.95895 40.34633 43.90681 #> 5: Chinstrap Dream 3200 46.21778 43.53954 45.90929 49.19173 #> 6: Gentoo Dream 3200 42.60465 39.89647 42.63520 46.28769 #> 7: Adelie Torgersen 3200 39.91652 36.80227 39.79806 43.68842 #> 8: Chinstrap Torgersen 3200 44.27807 41.95470 44.40742 46.68848 #> 9: Gentoo Torgersen 3200 42.09510 39.49863 41.80049 45.81833 #> 10: Adelie Biscoe 3550 40.77971 38.04027 40.59561 44.57505 #> 11: Chinstrap Biscoe 3550 45.81304 43.52102 45.73116 48.36366 #> 12: Gentoo Biscoe 3550 43.31233 40.77355 43.03077 47.22936 #> 13: Adelie Dream 3550 40.77741 38.07399 40.78175 44.37273 #> 14: Chinstrap Dream 3550 47.30926 44.80493 46.77540 50.47092 #> 15: Gentoo Dream 3550 43.26955 40.86119 43.16204 46.89190 #> 16: Adelie Torgersen 3550 40.25780 37.35251 40.07871 44.04576 #> 17: Chinstrap Torgersen 3550 44.77911 42.60161 44.81944 47.14986 #> 18: Gentoo Torgersen 3550 42.49520 39.95866 42.14160 46.26237 #> 19: Adelie Biscoe 3975 41.61744 38.94515 41.36634 45.38752 #> 20: Chinstrap Biscoe 3975 46.59363 44.59970 46.44923 49.11457 #> 21: Gentoo Biscoe 3975 44.07857 41.60792 43.74562 47.85109 #> 22: Adelie Dream 3975 41.50511 39.06187 41.24741 45.13027 #> 23: Chinstrap Dream 3975 48.14978 45.87390 47.54867 51.50683 #> 24: Gentoo Dream 3975 44.01928 41.70577 43.84099 47.50470 #> 25: Adelie Torgersen 3975 40.94764 38.12519 40.66759 44.73689 #> 26: Chinstrap Torgersen 3975 45.44820 43.49986 45.44036 47.63243 #> 27: Gentoo Torgersen 3975 43.13791 40.70628 42.70627 46.87306 #> 28: Adelie Biscoe 4700 42.93914 40.48463 42.44768 46.81756 #> 29: Chinstrap Biscoe 4700 47.18517 45.40866 47.07739 49.55747 #> 30: Gentoo Biscoe 4700 45.32541 43.08173 44.93498 49.23391 #> 31: Adelie Dream 4700 42.73806 40.44229 42.22226 46.49936 #> 32: Chinstrap Dream 4700 48.37278 46.34335 48.00781 51.18955 #> 33: Gentoo Dream 4700 45.09132 42.88328 44.79530 48.82180 #> 34: Adelie Torgersen 4700 42.09349 39.72074 41.56168 45.68838 #> 35: Chinstrap Torgersen 4700 46.16807 44.38410 46.09525 48.35127 #> 36: Gentoo Torgersen 4700 44.31621 42.18968 43.81773 47.98024 #> 37: Adelie Biscoe 5300 43.89769 41.43335 43.28504 48.10892 #> 38: Chinstrap Biscoe 5300 47.53721 45.66038 47.52770 49.88701 #> 39: Gentoo Biscoe 5300 46.16115 43.81722 45.59309 50.57469 #> 40: Adelie Dream 5300 43.59846 41.25825 43.24518 47.46193 #> 41: Chinstrap Dream 5300 48.48139 46.36282 48.25679 51.02996 #> 42: Gentoo Dream 5300 45.91819 43.62832 45.54110 49.91622 #> 43: Adelie Torgersen 5300 42.92879 40.66576 42.31072 46.76406 #> 44: Chinstrap Torgersen 5300 46.59576 44.80400 46.49196 49.03906 #> 45: Gentoo Torgersen 5300 45.11384 42.95190 44.51289 49.27629 #> species island body_mass_g mean lwr medn upr pd_new <- orsf_pd_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new #> variable value level mean lwr medn upr #> 1: species NA Adelie 41.90271 37.10417 41.51723 48.51478 #> 2: species NA Chinstrap 47.11314 42.40419 46.96478 51.51392 #> 3: species NA Gentoo 44.37038 39.87306 43.89889 51.21635 #> 4: island NA Biscoe 44.21332 37.22711 45.27862 51.21635 #> 5: island NA Dream 44.43354 37.01471 45.57261 51.51392 #> 6: island NA Torgersen 43.29539 37.01513 44.26924 49.84391 #> 7: body_mass_g 3200 42.84625 37.03978 43.95991 49.19173 #> 8: body_mass_g 3550 43.53326 37.56730 44.43756 50.47092 #> 9: body_mass_g 3975 44.30431 38.31567 45.22089 51.50683 #> 10: body_mass_g 4700 45.22525 39.88199 46.34680 51.18955 #> 11: body_mass_g 5300 45.91412 40.84742 46.95327 51.48851 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') pd_new <- orsf_pd_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) pd_new #> species island mean lwr medn upr #> 1: Adelie Biscoe 41.98024 37.22711 41.65252 48.51478"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"survival","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Survival","title":"PD and ICE curves with ORSF","text":"Begin fitting oblique survival random forest: Compute PD using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get PD value:","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) pd_train <- orsf_pd_inb(fit_surv, pred_spec = list(bili = 1:5)) pd_train #> pred_horizon bili mean lwr medn upr #> 1: 1826.25 1 0.2575450 0.02234786 0.1334170 0.8917942 #> 2: 1826.25 2 0.3130469 0.06853733 0.1906695 0.9203372 #> 3: 1826.25 3 0.3711963 0.11409793 0.2582027 0.9416791 #> 4: 1826.25 4 0.4248968 0.15648381 0.3334579 0.9591581 #> 5: 1826.25 5 0.4671699 0.20123406 0.3855137 0.9655296 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili)) pd_train #> pred_horizon bili mean lwr medn upr #> 1: 1826.25 0.590 0.2493753 0.02035041 0.1250263 0.8823385 #> 2: 1826.25 0.725 0.2517103 0.02060111 0.1281814 0.8836536 #> 3: 1826.25 1.500 0.2807082 0.03964900 0.1601715 0.9040617 #> 4: 1826.25 3.500 0.3968251 0.13431288 0.2934565 0.9501230 #> 5: 1826.25 7.210 0.5352155 0.27869513 0.4658256 0.9782084 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) pd_train #> pred_horizon bili mean lwr medn upr #> 1: 500 0.590 0.06217164 0.0004433990 0.008765301 0.5918852 #> 2: 1000 0.590 0.14282695 0.0057937418 0.056509484 0.7381953 #> 3: 1500 0.590 0.20944972 0.0136094784 0.092379507 0.8577223 #> 4: 2000 0.590 0.26917477 0.0230476894 0.146421502 0.8918696 #> 5: 2500 0.590 0.31901518 0.0631155452 0.203673185 0.9034059 #> 6: 3000 0.590 0.39244000 0.0911566314 0.302726475 0.9239494 #> 7: 500 0.725 0.06287876 0.0004462367 0.009001904 0.5980510 #> 8: 1000 0.725 0.14409310 0.0063321712 0.056833294 0.7448126 #> 9: 1500 0.725 0.21143724 0.0140736894 0.093685200 0.8597396 #> 10: 2000 0.725 0.27150368 0.0235448705 0.147022224 0.8940497 #> 11: 2500 0.725 0.32014805 0.0626303822 0.203946002 0.9073003 #> 12: 3000 0.725 0.39518173 0.0911457406 0.308428469 0.9252028 #> 13: 500 1.500 0.06712295 0.0012717884 0.011028398 0.6240769 #> 14: 1000 1.500 0.15802582 0.0114789623 0.068332010 0.7683888 #> 15: 1500 1.500 0.23407183 0.0287320952 0.117289745 0.8789647 #> 16: 2000 1.500 0.30235436 0.0467927208 0.180096425 0.9143235 #> 17: 2500 1.500 0.35354874 0.0845866747 0.238415966 0.9265099 #> 18: 3000 1.500 0.43604287 0.1311103304 0.348078730 0.9438196 #> 19: 500 3.500 0.08677320 0.0052087533 0.028244374 0.6741102 #> 20: 1000 3.500 0.22427808 0.0519179775 0.139857107 0.8277541 #> 21: 1500 3.500 0.32788654 0.0901983241 0.217982772 0.9371150 #> 22: 2000 3.500 0.41708208 0.1445328597 0.313224605 0.9566091 #> 23: 2500 3.500 0.49334883 0.2195110942 0.402932569 0.9636221 #> 24: 3000 3.500 0.56094391 0.2647541788 0.503509668 0.9734948 #> 25: 500 7.210 0.12591911 0.0220920570 0.063283130 0.7522611 #> 26: 1000 7.210 0.32642477 0.1353851175 0.259731888 0.8879218 #> 27: 1500 7.210 0.46409472 0.2181840827 0.387142510 0.9700903 #> 28: 2000 7.210 0.55116942 0.2912654769 0.484118150 0.9811496 #> 29: 2500 7.210 0.62008114 0.3709845684 0.568822502 0.9844945 #> 30: 3000 7.210 0.68030697 0.4247511750 0.646009789 0.9888637 #> pred_horizon bili mean lwr medn upr"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"one-variable-moving-horizon","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"One variable, moving horizon","title":"PD and ICE curves with ORSF","text":"next sections, update orsf_fit include data pbc_orsf instead just training sample: effect predictor varies time? Partial dependence can show . inspection, can see males higher risk females difference risk grows time. can also seen viewing ratio expected risk time: get view PD number variables training data, use orsf_summarize_uni(). function computes --bag PD important n_variables returns nicely formatted view output: ‘summary’ object can converted data.table downstream plotting tables.","code":"# a rare case of modify_in_place = TRUE orsf_update(fit_surv, data = pbc_orsf, modify_in_place = TRUE) fit_surv #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 500 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 21.038 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.84 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> ----------------------------------------- pd_sex_tv <- orsf_pd_oob(fit_surv, pred_spec = pred_spec_auto(sex), pred_horizon = seq(365, 365*5)) ggplot(pd_sex_tv) + aes(x = pred_horizon, y = mean, color = sex) + geom_line() + labs(x = 'Time since baseline', y = 'Expected risk') library(data.table) ratio_tv <- pd_sex_tv[ , .(ratio = mean[sex == 'm'] / mean[sex == 'f']), by = pred_horizon ] ggplot(ratio_tv, aes(x = pred_horizon, y = ratio)) + geom_line(color = 'grey') + geom_smooth(color = 'black', se = FALSE) + labs(x = 'time since baseline', y = 'ratio in expected risk for males versus females') pd_smry <- orsf_summarize_uni(fit_surv, n_variables = 4) pd_smry #> #> -- ascites (VI Rank: 1) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3083328 0.1985589 0.06581247 0.5241336 #> 1 0.4702396 0.3975953 0.27481738 0.6564321 #> #> -- bili (VI Rank: 2) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2356543 0.1536301 0.05872720 0.3719578 #> 0.80 0.2398021 0.1609720 0.06167673 0.3776136 #> 1.40 0.2613612 0.1809950 0.07893386 0.4064484 #> 3.52 0.3702763 0.3118827 0.17050712 0.5447088 #> 7.25 0.4780580 0.4406202 0.29442977 0.6434075 #> #> -- edema (VI Rank: 3) -------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3035731 0.1840849 0.06509174 0.5228237 #> 0.5 0.3558716 0.2649457 0.11132293 0.5831396 #> 1 0.4693915 0.3961470 0.28211662 0.6331870 #> #> -- copper (VI Rank: 4) ------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 25.5 0.2632768 0.1622871 0.05581251 0.4308234 #> 42.8 0.2707739 0.1703028 0.05887747 0.4418590 #> 74.0 0.2908707 0.1940176 0.07155433 0.4768302 #> 129 0.3444258 0.2651729 0.11918406 0.5574967 #> 214 0.4245218 0.3577346 0.21408331 0.6238041 #> #> Predicted risk at time t = 1826.25 for top 4 predictors head(as.data.table(pd_smry)) #> variable importance Value Mean Median 25th % 75th % #> 1: ascites 0.4960630 0 0.3083328 0.1985589 0.06581247 0.5241336 #> 2: ascites 0.4960630 1 0.4702396 0.3975953 0.27481738 0.6564321 #> 3: bili 0.4160074 0.60 0.2356543 0.1536301 0.05872720 0.3719578 #> 4: bili 0.4160074 0.80 0.2398021 0.1609720 0.06167673 0.3776136 #> 5: bili 0.4160074 1.40 0.2613612 0.1809950 0.07893386 0.4064484 #> 6: bili 0.4160074 3.52 0.3702763 0.3118827 0.17050712 0.5447088 #> pred_horizon level #> 1: 1826.25 0 #> 2: 1826.25 1 #> 3: 1826.25 #> 4: 1826.25 #> 5: 1826.25 #> 6: 1826.25 "},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"multiple-variables-jointly","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Multiple variables, jointly","title":"PD and ICE curves with ORSF","text":"Partial dependence can show expected value model’s predictions function specific predictor, function multiple predictors. instance, can estimate predicted risk joint function bili, edema, trt: inspection, model’s predictions indicate slightly lower risk placebo group, seem change much different values bili edema. clear increase predicted risk higher levels edema higher levels bili slope predicted risk function bili appears highest among patients edema 0.5. effect bili modified edema 0.5? quick sanity check coxph suggests .","code":"pred_spec = pred_spec_auto(bili, edema, trt) pd_bili_edema <- orsf_pd_oob(fit_surv, pred_spec) ggplot(pd_bili_edema) + aes(x = bili, y = medn, col = trt, linetype = edema) + geom_line() + labs(y = 'Expected predicted risk') library(survival) pbc_orsf$edema_05 <- ifelse(pbc_orsf$edema == '0.5', 'yes', 'no') fit_cph <- coxph(Surv(time,status) ~ edema_05 * bili, data = pbc_orsf) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> edema_05 -546.83 6.7248 1 0.009508 ** #> bili -513.59 66.4689 1 3.555e-16 *** #> edema_05:bili -510.54 6.1112 1 0.013433 * #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"find-interactions-using-pd","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Find interactions using PD","title":"PD and ICE curves with ORSF","text":"Random forests good using interactions, less good telling . Use orsf_vint() apply method variable interaction scoring PD described Greenwell et al (2018). can take little lots predictors. use sanity check coxph see interactions detected using standard test: Note: Caution warranted interpreting statistical hypotheses motivated data tested . Results like p-values interaction shown interpreted exploratory.","code":"vint_scores <- orsf_vint(fit_surv) vint_scores[1:5] #> interaction score #> 1: albumin..protime 1.1597307 #> 2: protime..stage 1.0710384 #> 3: copper..protime 0.7958742 #> 4: albumin..stage 0.7672757 #> 5: bili..stage 0.7619507 fit_cph <- coxph(Surv(time,status) ~ albumin * protime, data = pbc_orsf) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> albumin -526.29 47.801 1 4.717e-12 *** #> protime -514.89 22.806 1 1.792e-06 *** #> albumin:protime -511.76 6.252 1 0.01241 * #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 fit_cph <- update(fit_cph, . ~ stage * protime) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> stage -528.61 43.1626 3 2.273e-09 *** #> protime -516.05 25.1244 1 5.375e-07 *** #> stage:protime -514.06 3.9822 3 0.2634 #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 fit_cph <- update(fit_cph, . ~ copper * protime) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> copper -524.72 50.9444 1 9.502e-13 *** #> protime -513.02 23.3872 1 1.325e-06 *** #> copper:protime -508.86 8.3216 1 0.003918 ** #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"individual-conditional-expectations-ice","dir":"Articles","previous_headings":"","what":"Individual conditional expectations (ICE)","title":"PD and ICE curves with ORSF","text":"Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"classification-1","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Classification","title":"PD and ICE curves with ORSF","text":"Compute ICE using --bag data flipper_length_mm = c(190, 210). two identifiers output: id_variable identifier current value variable(s) data. redundant one variable, helpful multiple variables. id_row identifier observation original data. Note predicted probabilities returned class observation data. Predicted probabilities given observation given variable value sum 1. example,","code":"pred_spec <- list(flipper_length_mm = c(190, 210)) ice_oob <- orsf_ice_oob(fit_clsf, pred_spec = pred_spec) ice_oob #> id_variable id_row class flipper_length_mm pred #> 1: 1 1 Adelie 190 0.92045213 #> 2: 1 2 Adelie 190 0.80427932 #> 3: 1 3 Adelie 190 0.84342550 #> 4: 1 4 Adelie 190 0.93514694 #> 5: 1 5 Adelie 190 0.97172229 #> --- #> 896: 2 146 Gentoo 210 0.25779089 #> 897: 2 147 Gentoo 210 0.04806888 #> 898: 2 148 Gentoo 210 0.07926342 #> 899: 2 149 Gentoo 210 0.84597108 #> 900: 2 150 Gentoo 210 0.10191162 ice_oob %>% .[flipper_length_mm == 190] %>% .[id_row == 1] %>% .[['pred']] %>% sum() #> [1] 1"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"regression-1","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Regression","title":"PD and ICE curves with ORSF","text":"Compute ICE using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"pred_spec <- list(flipper_length_mm = c(190, 210)) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row flipper_length_mm pred #> 1: 1 1 190 37.94483 #> 2: 1 2 190 37.61595 #> 3: 1 3 190 37.53681 #> 4: 1 4 190 39.49476 #> 5: 1 5 190 38.95635 #> --- #> 362: 2 179 210 51.80471 #> 363: 2 180 210 47.27183 #> 364: 2 181 210 47.05031 #> 365: 2 182 210 50.39028 #> 366: 2 183 210 48.44774 pred_spec = pred_spec_auto(species, island, body_mass_g) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row species island body_mass_g pred #> 1: 1 1 Adelie Biscoe 3200 37.78339 #> 2: 1 2 Adelie Biscoe 3200 37.73273 #> 3: 1 3 Adelie Biscoe 3200 37.71248 #> 4: 1 4 Adelie Biscoe 3200 40.25782 #> 5: 1 5 Adelie Biscoe 3200 40.04074 #> --- #> 8231: 45 179 Gentoo Torgersen 5300 46.14559 #> 8232: 45 180 Gentoo Torgersen 5300 43.98050 #> 8233: 45 181 Gentoo Torgersen 5300 44.59837 #> 8234: 45 182 Gentoo Torgersen 5300 44.85146 #> 8235: 45 183 Gentoo Torgersen 5300 44.23710 ice_new <- orsf_ice_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row variable value level pred #> 1: 1 1 species NA Adelie 37.74136 #> 2: 1 2 species NA Adelie 37.42367 #> 3: 1 3 species NA Adelie 37.04598 #> 4: 1 4 species NA Adelie 39.89602 #> 5: 1 5 species NA Adelie 39.14848 #> --- #> 2009: 5 179 body_mass_g 5300 51.50196 #> 2010: 5 180 body_mass_g 5300 47.27055 #> 2011: 5 181 body_mass_g 5300 48.34064 #> 2012: 5 182 body_mass_g 5300 48.75828 #> 2013: 5 183 body_mass_g 5300 48.11020 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') ice_new <- orsf_ice_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row species island pred #> 1: 1 1 Adelie Biscoe 38.52327 #> 2: 1 2 Adelie Biscoe 38.32073 #> 3: 1 3 Adelie Biscoe 37.71248 #> 4: 1 4 Adelie Biscoe 41.68380 #> 5: 1 5 Adelie Biscoe 40.91140 #> --- #> 179: 1 179 Adelie Biscoe 43.09493 #> 180: 1 180 Adelie Biscoe 38.79455 #> 181: 1 181 Adelie Biscoe 39.37734 #> 182: 1 182 Adelie Biscoe 40.71952 #> 183: 1 183 Adelie Biscoe 39.34501"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"survival-1","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Survival","title":"PD and ICE curves with ORSF","text":"Compute ICE using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get ICE value: Multi-prediction horizon ice comes minimal extra computational cost. Use fine grid time values assess whether predictors time-varying effects.","code":"ice_train <- orsf_ice_inb(fit_surv, pred_spec = list(bili = 1:5)) ice_train #> id_variable id_row pred_horizon bili pred #> 1: 1 1 1826.25 1 0.9015162 #> 2: 1 2 1826.25 1 0.1019426 #> 3: 1 3 1826.25 1 0.6821646 #> 4: 1 4 1826.25 1 0.3623411 #> 5: 1 5 1826.25 1 0.1374271 #> --- #> 1376: 5 272 1826.25 5 0.2650957 #> 1377: 5 273 1826.25 5 0.3065318 #> 1378: 5 274 1826.25 5 0.3503776 #> 1379: 5 275 1826.25 5 0.1652897 #> 1380: 5 276 1826.25 5 0.3549165 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili)) ice_train #> id_variable id_row pred_horizon bili pred #> 1: 1 1 1826.25 0.60 0.89210440 #> 2: 1 2 1826.25 0.60 0.09186876 #> 3: 1 3 1826.25 0.60 0.65503431 #> 4: 1 4 1826.25 0.60 0.34622748 #> 5: 1 5 1826.25 0.60 0.13310425 #> --- #> 1376: 5 272 1826.25 7.25 0.31258148 #> 1377: 5 273 1826.25 7.25 0.35478676 #> 1378: 5 274 1826.25 7.25 0.41559176 #> 1379: 5 275 1826.25 7.25 0.25301890 #> 1380: 5 276 1826.25 7.25 0.44533769 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) ice_train #> id_variable id_row pred_horizon bili pred #> 1: 1 1 500 0.60 0.5949598 #> 2: 1 1 1000 0.60 0.7652137 #> 3: 1 1 1500 0.60 0.8751746 #> 4: 1 1 2000 0.60 0.9057135 #> 5: 1 1 2500 0.60 0.9231915 #> --- #> 8276: 5 276 1000 7.25 0.2111306 #> 8277: 5 276 1500 7.25 0.3642278 #> 8278: 5 276 2000 7.25 0.4850492 #> 8279: 5 276 2500 7.25 0.5720362 #> 8280: 5 276 3000 7.25 0.6206786"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"visualizing-ice-curves","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Visualizing ICE curves","title":"PD and ICE curves with ORSF","text":"Inspecting ICE curves observation can help identify whether heterogeneity model’s predictions. .e., effect variable follow pattern data, groups variable impacts risk differently? going turn boundary checking orsf_ice_oob setting boundary_checks = FALSE, allow generate ICE curves go beyond 90th percentile bili. plots, helpful scale ICE data. subtract initial value predicted risk (.e., bili = 1) observation’s conditional expectation values. , Every curve start 0 plot shows change predicted risk function bili. Now can visualize curves. inspection figure, individual slopes cluster around overall trend - Good! small number individual slopes appear flat. may helpful investigate .","code":"pred_spec <- list(bili = seq(1, 10, length.out = 25)) ice_oob <- orsf_ice_oob(fit_surv, pred_spec, boundary_checks = FALSE) ice_oob #> id_variable id_row pred_horizon bili pred #> 1: 1 1 1826.25 1 0.8790861 #> 2: 1 2 1826.25 1 0.8132035 #> 3: 1 3 1826.25 1 0.6240238 #> 4: 1 4 1826.25 1 0.7461603 #> 5: 1 5 1826.25 1 0.5754091 #> --- #> 6896: 25 272 1826.25 10 0.7018976 #> 6897: 25 273 1826.25 10 0.4606246 #> 6898: 25 274 1826.25 10 0.3347082 #> 6899: 25 275 1826.25 10 0.6046024 #> 6900: 25 276 1826.25 10 0.2789017 ice_oob[, pred_subtract := rep(pred[id_variable==1], times=25)] ice_oob[, pred := pred - pred_subtract] ggplot(ice_oob, aes(x = bili, y = pred, group = id_row)) + geom_line(alpha = 0.15) + labs(y = 'Change in predicted risk') + geom_smooth(se = FALSE, aes(group = 1))"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"limitations-of-pd","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Limitations of PD","title":"PD and ICE curves with ORSF","text":"Partial dependence number known limitations assumptions users aware (see Hooker, 2021). particular, partial dependence less intuitive >2 predictors examined jointly, assumed feature(s) partial dependence computed correlated features (likely true many cases). Accumulated local effect plots can used (see ) case feature independence valid assumption.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"references","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"References","title":"PD and ICE curves with ORSF","text":"Hooker, Giles, Mentch, Lucas, Zhou, Siyu (2021). “Unrestricted permutation forces extrapolation: variable importance requires least one model, free variable importance.” Statistics Computing, 31, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/authors.html","id":null,"dir":"","previous_headings":"","what":"Authors","title":"Authors and Citation","text":"Byron Jaeger. Author, maintainer. Nicholas Pajewski. Contributor. Sawyer Welden. Contributor. Christopher Jackson. Reviewer. Marvin Wright. Reviewer. Lukas Burk. Reviewer.","code":""},{"path":"https://bcjaeger.github.io/aorsf/authors.html","id":"citation","dir":"","previous_headings":"","what":"Citation","title":"Authors and Citation","text":"Jaeger et al. (2022). aorsf: R package supervised learning using oblique random survival forest. Journal Open Source Software, 7(77), 4705. https://doi.org/10.21105/joss.04705. Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM. Accelerated interpretable oblique random survival forests. Journal Computational Graphical Statistics. 2023 Aug 3:1-6. Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min YI, Mcclure LA, Howard G, Simon N. Oblique Random Survival Forests. Annals Applied Statistics. 13(3): 1847-1883. URL https://doi.org/10.1214/19-AOAS1261 DOI: 10.1214/19-AOAS1261","code":"@Article{, title = {aorsf: An R package for supervised learning using the oblique random survival forest}, author = {Byron C. Jaeger and Sawyer Welden and Kristin Lenoir and Nicholas M. Pajewski}, journal = {Journal of Open Source Software}, year = {2022}, volume = {7}, number = {77}, pages = {4705}, url = {https://doi.org/10.21105/joss.04705}, } @Article{, title = {Accelerated and interpretable oblique random survival forests}, author = {Byron C. Jaeger and Sawyer Welden and Kristin Lenoir and Jaime L. Speiser and Matthew W. Segar and Ambarish Pandey and Nicholas M. Pajewski}, journal = {Journal of Computational and Graphical Statistics}, year = {2023}, url = {https://doi.org/10.1080/10618600.2023.2231048}, } @Article{, title = {Oblique Random Survival Forests}, author = {Byron C. Jaeger and D. Leann Long and Dustin M. Long and Mario Sims and Jeff M. Szychowski and Yuan-I Min and Leslie A. Mcclure and George Howard and Noah Simon}, journal = {Annals of Applied Statistics}, year = {2019}, volume = {13}, number = {3}, pages = {1847--1883}, url = {https://doi.org/10.1214/19-AOAS1261}, }"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"aorsf-","dir":"","previous_headings":"","what":"Accelerated Oblique Random Forests","title":"Accelerated Oblique Random Forests","text":"Fit, interpret, make predictions oblique random forests (RFs).","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"why-aorsf","dir":"","previous_headings":"","what":"Why aorsf?","title":"Accelerated Oblique Random Forests","text":"Fast versatile tools oblique RFs.1 Accurate predictions.2 Intuitive design formula based interface. Extensive input checks informative error messages. Compatible tidymodels mlr3","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"installation","dir":"","previous_headings":"","what":"Installation","title":"Accelerated Oblique Random Forests","text":"can install aorsf CRAN using can install development version aorsf GitHub :","code":"install.packages(\"aorsf\") # install.packages(\"remotes\") remotes::install_github(\"ropensci/aorsf\")"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"get-started","dir":"","previous_headings":"","what":"Get started","title":"Accelerated Oblique Random Forests","text":"aorsf fits several types oblique RFs orsf() function, including classification, regression, survival RFs. classification, fit oblique RF predict penguin species using penguin data magnificent palmerpenguins R package regression, use data predict bill length penguins: personal favorite oblique survival RF accelerated Cox regression first type oblique RF aorsf provided (see JCGS paper). , use predict mortality risk following diagnosis primary biliary cirrhosis:","code":"library(aorsf) library(tidyverse) # An oblique classification RF penguin_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = species ~ .) penguin_fit #> ---------- Oblique random classification forest #> #> Linear combinations: Accelerated Logistic regression #> N observations: 333 #> N classes: 3 #> N trees: 5 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 6 #> Min observations in leaf: 5 #> OOB stat value: 0.99 #> OOB stat type: AUC-ROC #> Variable importance: anova #> #> ----------------------------------------- # An oblique regression RF bill_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = bill_length_mm ~ .) bill_fit #> ---------- Oblique random regression forest #> #> Linear combinations: Accelerated Linear regression #> N observations: 333 #> N trees: 5 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 42.6 #> Min observations in leaf: 5 #> OOB stat value: 0.76 #> OOB stat type: RSQ #> Variable importance: anova #> #> ----------------------------------------- # An oblique survival RF pbc_fit <- orsf(data = pbc_orsf, n_tree = 5, formula = Surv(time, status) ~ . - id) pbc_fit #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 20.4 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.79 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"what-does-oblique-mean","dir":"","previous_headings":"","what":"What does “oblique” mean?","title":"Accelerated Oblique Random Forests","text":"Decision trees grown splitting set training data non-overlapping subsets, goal similarity within new subsets . subsets created single predictor, decision tree axis-based subset boundaries perpendicular axis predictor. linear combinations (.e., weighted sum) variables used instead single variable, tree oblique boundaries neither parallel perpendicular axis. Figure: Decision trees classification axis-based splitting (left) oblique splitting (right). Cases orange squares; controls purple circles. trees partition predictor space defined variables X1 X2, oblique splits better job separating two classes. , difference translate real data, impact random forests comprising hundreds axis-based oblique trees? demonstrate using penguin data.3 also use function make several plots: also use grid points plotting decision surfaces: use orsf mtry=1 fit axis-based trees: Next use orsf_update copy modify original model, expanding fit oblique tree using mtry=2 instead mtry=1, include 500 trees instead 1: now need visualize decision surfaces using predictions four fits: Figure: Axis-based oblique decision surfaces single tree ensemble 500 trees. Axis-based trees boundaries perpendicular predictor axes, whereas oblique trees can boundaries neither parallel perpendicular predictor axes. Axis-based forests tend ‘step-function’ decision boundaries, oblique forests tend smooth decision boundaries.","code":"plot_decision_surface <- function(predictions, title, grid){ # this is not a general function for plotting # decision surfaces. It just helps to minimize # copying and pasting of code. class_preds <- bind_cols(grid, predictions) %>% pivot_longer(cols = c(Adelie, Chinstrap, Gentoo)) %>% group_by(flipper_length_mm, bill_length_mm) %>% arrange(desc(value)) %>% slice(1) cols <- c(\"darkorange\", \"purple\", \"cyan4\") ggplot(class_preds, aes(bill_length_mm, flipper_length_mm)) + geom_contour_filled(aes(z = value, fill = name), alpha = .25) + geom_point(data = penguins_orsf, aes(color = species, shape = species), alpha = 0.5) + scale_color_manual(values = cols) + scale_fill_manual(values = cols) + labs(x = \"Bill length, mm\", y = \"Flipper length, mm\") + theme_minimal() + scale_x_continuous(expand = c(0,0)) + scale_y_continuous(expand = c(0,0)) + theme(panel.grid = element_blank(), panel.border = element_rect(fill = NA), legend.position = '') + labs(title = title) } grid <- expand_grid( flipper_length_mm = seq(min(penguins_orsf$flipper_length_mm), max(penguins_orsf$flipper_length_mm), len = 200), bill_length_mm = seq(min(penguins_orsf$bill_length_mm), max(penguins_orsf$bill_length_mm), len = 200) ) fit_axis_tree <- penguins_orsf %>% orsf(species ~ bill_length_mm + flipper_length_mm, n_tree = 1, mtry = 1, tree_seeds = 106760) fit_axis_forest <- fit_axis_tree %>% orsf_update(n_tree = 500) fit_oblique_tree <- fit_axis_tree %>% orsf_update(mtry = 2) fit_oblique_forest <- fit_oblique_tree %>% orsf_update(n_tree = 500) preds <- list(fit_axis_tree, fit_axis_forest, fit_oblique_tree, fit_oblique_forest) %>% map(predict, new_data = grid, pred_type = 'prob') titles <- c(\"Axis-based tree\", \"Axis-based forest\", \"Oblique tree\", \"Oblique forest\") plots <- map2(preds, titles, plot_decision_surface, grid = grid)"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"variable-importance","dir":"","previous_headings":"","what":"Variable importance","title":"Accelerated Oblique Random Forests","text":"importance individual predictor variables can estimated three ways using aorsf can used type oblique RF. Also, variable importance functions always return named character vector negation2: variable assessed separately multiplying variable’s coefficients -1 determining much model’s performance changes. worse model’s performance negating coefficients given variable, important variable. technique promising b/c require permutation emphasizes variables larger coefficients linear combinations, also relatively new hasn’t studied much permutation importance. See Jaeger, (2023) details technique. permutation: variable assessed separately randomly permuting variable’s values determining much model’s performance changes. worse model’s performance permuting values given variable, important variable. technique flexible, intuitive, frequently used. also several known limitations analysis variance (ANOVA)4: p-value computed coefficient linear combination variables decision tree. Importance individual predictor variable proportion times p-value coefficient < 0.01. technique efficient computationally, may effective permutation negation terms selecting signal noise variables. See Menze, 2011 details technique. can supply R function estimate --bag error (see oob vignette) estimate --bag variable importance (see orsf_vi examples)","code":"orsf_vi_negate(pbc_fit) #> bili copper stage sex age #> 0.1552460736 0.1156218837 0.0796917628 0.0533427094 0.0283132385 #> albumin trt chol alk.phos platelet #> 0.0279823814 0.0168238416 0.0153010749 0.0148718669 0.0094582765 #> edema ascites spiders protime hepato #> 0.0067975986 0.0065505801 0.0062356214 -0.0004653046 -0.0026664147 #> ast trig #> -0.0028902524 -0.0106616501 orsf_vi_permute(penguin_fit) #> bill_length_mm bill_depth_mm body_mass_g island #> 0.121351910 0.101846889 0.097822451 0.080772909 #> sex flipper_length_mm year #> 0.035053517 0.008270751 -0.008058339 orsf_vi_anova(bill_fit) #> species sex bill_depth_mm flipper_length_mm #> 0.51652893 0.27906977 0.06315789 0.04950495 #> body_mass_g island year #> 0.04807692 0.02687148 0.00000000"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"partial-dependence-pd","dir":"","previous_headings":"","what":"Partial dependence (PD)","title":"Accelerated Oblique Random Forests","text":"Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model’s prediction.. can use specific values predictor compute PD let aorsf pick reasonable values use pred_spec_auto(): summary function, orsf_summarize_uni(), computes PD many variables ask , using sensible values. PD, see vignette","code":"# pick your own values orsf_pd_oob(bill_fit, pred_spec = list(species = c(\"Adelie\", \"Gentoo\"))) #> species mean lwr medn upr #> #> 1: Adelie 39.99394 35.76532 39.80782 46.13931 #> 2: Gentoo 46.66565 40.02938 46.88517 51.61367 # let aorsf pick reasonable values for you: orsf_pd_oob(bill_fit, pred_spec = pred_spec_auto(bill_depth_mm, island)) #> bill_depth_mm island mean lwr medn upr #> #> 1: 14.3 Biscoe 43.94960 35.90421 45.30159 51.05109 #> 2: 15.6 Biscoe 44.24705 36.62759 45.57321 51.08020 #> 3: 17.3 Biscoe 44.84757 36.53804 45.62910 53.93833 #> 4: 18.7 Biscoe 45.08939 36.35893 46.16893 54.42075 #> 5: 19.5 Biscoe 45.13608 36.21033 46.08023 54.42075 #> --- #> 11: 14.3 Torgersen 43.55984 35.47143 44.18127 51.05109 #> 12: 15.6 Torgersen 43.77317 35.44683 44.28406 51.08020 #> 13: 17.3 Torgersen 44.56465 35.84585 44.83694 53.93833 #> 14: 18.7 Torgersen 44.68367 35.44010 44.86667 54.42075 #> 15: 19.5 Torgersen 44.64605 35.44010 44.86667 54.42075 orsf_summarize_uni(pbc_fit, n_variables = 2) #> #> -- bili (VI Rank: 1) ----------------------------- #> #> |----------------- Risk -----------------| #> Value Mean Median 25th % 75th % #> #> 0.60 0.2098108 0.07168855 0.01138461 0.2860450 #> 0.80 0.2117933 0.07692308 0.01709469 0.2884990 #> 1.40 0.2326560 0.08445419 0.02100837 0.3563622 #> 3.55 0.4265979 0.35820106 0.05128824 0.7342923 #> 7.30 0.4724608 0.44746241 0.11759259 0.8039683 #> #> -- copper (VI Rank: 2) --------------------------- #> #> |----------------- Risk -----------------| #> Value Mean Median 25th % 75th % #> #> 25.0 0.2332412 0.04425936 0.01587919 0.3888304 #> 42.5 0.2535448 0.07417582 0.01754386 0.4151786 #> 74.0 0.2825471 0.11111111 0.01988069 0.4770833 #> 130 0.3259604 0.18771003 0.04658385 0.5054348 #> 217 0.4213303 0.28571429 0.13345865 0.6859423 #> #> Predicted risk at time t = 1788 for top 2 predictors"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"individual-conditional-expectations-ice","dir":"","previous_headings":"","what":"Individual conditional expectations (ICE)","title":"Accelerated Oblique Random Forests","text":"Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor. ICE, see vignette","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"interaction-scores","dir":"","previous_headings":"","what":"Interaction scores","title":"Accelerated Oblique Random Forests","text":"orsf_vint() function computes score possible interaction model based PD using method described Greenwell et al, 2018.5 can slow larger datasets, substantial speedups occur making use multi-threading restricting search smaller set predictors. values score mean? values average standard deviation standard deviation PD one variable conditional variable. interpreted relative one another, .e., higher scoring interaction likely reflect real interaction two variables lower scoring one. interaction scores make sense? Let’s test top scoring lowest scoring interactions using coxph(). Note: exploratory true null hypothesis test. ? used data generate test null hypothesis. much conducting statistical inference test interactions coxph demonstrating interaction scores orsf_vint() provides consistent tests models.","code":"preds_interaction <- c(\"albumin\", \"protime\", \"bili\", \"spiders\", \"trt\") # While it is tempting to speed up `orsf_vint()` by growing a smaller # number of trees, results may become unstable with this shortcut. pbc_interactions <- pbc_fit %>% orsf_update(n_tree = 500, tree_seeds = 329) %>% orsf_vint(n_thread = 0, predictors = preds_interaction) pbc_interactions #> interaction score #> #> 1: albumin..protime 0.97837184 #> 2: protime..bili 0.78999788 #> 3: albumin..bili 0.59128756 #> 4: bili..spiders 0.13192184 #> 5: bili..trt 0.13192184 #> 6: albumin..spiders 0.06578222 #> 7: albumin..trt 0.06578222 #> 8: protime..spiders 0.03012718 #> 9: protime..trt 0.03012718 #> 10: spiders..trt 0.00000000 library(survival) # the top scoring interaction should get a lower p-value anova(coxph(Surv(time, status) ~ protime * albumin, data = pbc_orsf)) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> protime -538.51 23.353 1 1.349e-06 *** #> albumin -514.89 47.255 1 6.234e-12 *** #> protime:albumin -511.76 6.252 1 0.01241 * #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 # the bottom scoring interaction should get a higher p-value anova(coxph(Surv(time, status) ~ spiders * trt, data = pbc_orsf)) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> spiders -538.58 23.2159 1 1.448e-06 *** #> trt -538.39 0.3877 1 0.5335 #> spiders:trt -538.29 0.2066 1 0.6494 #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"comparison-to-existing-software","dir":"","previous_headings":"","what":"Comparison to existing software","title":"Accelerated Oblique Random Forests","text":"survival analysis, comparisons aorsf existing software presented JCGS paper. paper: describes aorsf detail summary procedures used tree fitting algorithm runs general benchmark comparing aorsf obliqueRSF several learners reports prediction accuracy computational efficiency learners. runs simulation study comparing variable importance techniques oblique survival RFs, axis based survival RFs, boosted trees. reports probability variable importance technique rank relevant variable higher importance irrelevant variable.","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"references","dir":"","previous_headings":"","what":"References","title":"Accelerated Oblique Random Forests","text":"Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min Y, Mcclure LA, Howard G, Simon N (2019). “Oblique random survival forests.” Annals Applied Statistics, 13(3). doi:10.1214/19-aoas1261 https://doi.org/10.1214/19-aoas1261. Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM (2023). “Accelerated interpretable oblique random survival forests.” Journal Computational Graphical Statistics, 1-16. doi:10.1080/10618600.2023.2231048 https://doi.org/10.1080/10618600.2023.2231048. Horst , Hill AP, Gorman KB (2020). palmerpenguins: Palmer Archipelago (Antarctica) penguin data. R package version 0.1.0, https://allisonhorst.github.io/palmerpenguins/. Menze, H B, Kelm, Michael B, Splitthoff, N D, Koethe, Ullrich, Hamprecht, F (2011). “oblique random forests.” Machine Learning Knowledge Discovery Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011, Proceedings, Part II 22, 453-469. Springer. Greenwell, M B, Boehmke, C B, McCarthy, J (2018). “simple effective model-based variable importance measure.” arXiv preprint arXiv:1805.04755.","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"funding","dir":"","previous_headings":"","what":"Funding","title":"Accelerated Oblique Random Forests","text":"developers aorsf received financial support Center Biomedical Informatics, Wake Forest University School Medicine. also received support National Center Advancing Translational Sciences National Institutes Health Award Number UL1TR001420. content solely responsibility authors necessarily represent official views National Institutes Health.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/aorsf-package.html","id":null,"dir":"Reference","previous_headings":"","what":"aorsf: Accelerated Oblique Random Forests — aorsf-package","title":"aorsf: Accelerated Oblique Random Forests — aorsf-package","text":"Fit, interpret, compute predictions oblique random forests. Includes support partial dependence, variable importance, passing customized functions variable importance identification linear combinations features. Methods oblique random survival forest described Jaeger et al., (2023) doi:10.1080/10618600.2023.2231048 .","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/aorsf-package.html","id":"author","dir":"Reference","previous_headings":"","what":"Author","title":"aorsf: Accelerated Oblique Random Forests — aorsf-package","text":"Maintainer: Byron Jaeger bjaeger@wakehealth.edu (ORCID) contributors: Nicholas Pajewski [contributor] Sawyer Welden swelden@wakehealth.edu [contributor] Christopher Jackson chris.jackson@mrc-bsu.cam.ac.uk [reviewer] Marvin Wright [reviewer] Lukas Burk [reviewer]","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":null,"dir":"Reference","previous_headings":"","what":"Coerce to data.table — as.data.table.orsf_summary_uni","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"Convert 'orsf_summary' object data.table object.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"","code":"# S3 method for orsf_summary_uni as.data.table(x, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"x object class 'orsf_summary_uni' ... used","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"data.table","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"","code":"if (FALSE) { library(data.table) object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 25) smry <- orsf_summarize_uni(object, n_variables = 2) as.data.table(smry) }"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":null,"dir":"Reference","previous_headings":"","what":"Oblique Random Forests — orsf","title":"Oblique Random Forests — orsf","text":"Grow specify oblique random forest. name orsf() implies function works survival forests, can used classification, regression, survival forests.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Oblique Random Forests — orsf","text":"","code":"orsf( data, formula, control = NULL, weights = NULL, n_tree = 500, n_split = 5, n_retry = 3, n_thread = 0, mtry = NULL, sample_with_replacement = TRUE, sample_fraction = 0.632, leaf_min_events = 1, leaf_min_obs = 5, split_rule = NULL, split_min_events = 5, split_min_obs = 10, split_min_stat = NULL, oobag_pred_type = NULL, oobag_pred_horizon = NULL, oobag_eval_every = NULL, oobag_fun = NULL, importance = \"anova\", importance_max_pvalue = 0.01, group_factors = TRUE, tree_seeds = NULL, attach_data = TRUE, no_fit = FALSE, na_action = \"fail\", verbose_progress = FALSE, ... ) orsf_train(object, attach_data = TRUE)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Oblique Random Forests — orsf","text":"data data.frame, tibble, data.table contains relevant variables. formula (formula) Two sided formula single outcome. terms right names predictor variables, symbol '.' may used indicate variables data except response. symbol '-' may also used indicate removal predictor. Details response vary depending forest type: Classification: response single variable, variable type factor data. Regression: response single variable, variable typee double integer least 10 unique numeric values data. Survival: response include time variable, followed status variable, may written inside call Surv (see examples). control (orsf_control) object returned one orsf_control functions: orsf_control_survival, orsf_control_classification, orsf_control_regression. NULL (default) use accelerated control, fastest available option. survival classification, Cox Logistic regression 1 iteration, regression ordinary least squares. weights (numeric vector) Optional. given, input length equal nrow(data) complete imputed data length equal nrow(na.omit(data)) na_action \"omit\". weights vector used count observations events prior growing node tree, orsf() scales weights sum(weights) == nrow(data). helps make tree depth consistent weighted un-weighted fits. n_tree (integer) number trees grow. Default n_tree = 500. n_split (integer) number cut-points assessed splitting node decision trees. Default n_split = 5. n_retry (integer) node splittable, current linear combination inputs unable provide valid split, orsf try new linear combination based different set randomly selected predictors, n_retry times. Default n_retry = 3. Set n_retry = 0 prevent retries. n_thread (integer) number threads use growing trees, computing predictions, computing importance. Default 0, allows suitable number threads used based availability. mtry (integer) Number predictors randomly included candidates splitting node. default smallest integer greater square root number total predictors, .e., mtry = ceiling(sqrt(number predictors)) sample_with_replacement (logical) TRUE (default), observations sampled replacement -bag sample created decision tree. FALSE, observations sampled without replacement tree -bag sample containing sample_fraction% original sample. sample_fraction (double) proportion observations trees' -bag sample contain, relative number rows data. used sample_with_replacement FALSE. Default value 0.632. leaf_min_events (integer) input relevant survival analysis, specifies minimum number events leaf node. Default leaf_min_events = 1 leaf_min_obs (integer) minimum number observations leaf node. Default leaf_min_obs = 5. split_rule (character) assess quality potential splitting rule node. Valid options survival : 'logrank' : log-rank test statistic (default). 'cstat' : Harrell's concordance statistic. classification, valid options : 'gini' : gini impurity (default) 'cstat' : area underneath ROC curve (AUC-ROC) regression, valid options : 'variance' : variance reduction (default) split_min_events (integer) minimum number events required node consider splitting . Default split_min_events = 5. input relevant survival trees. split_min_obs (integer) minimum number observations required node consider splitting . Default split_min_obs = 10. split_min_stat (double) minimum test statistic required split node. splits found statistic exceeding split_min_stat, given node either becomes leaf retry occurs (n_retry retries). Defaults 3.84 split_rule = 'logrank' 0.55 split_rule = 'cstat' (see first note ) 0.00 split_rule = 'gini' (see second note ) 0.00 split_rule = 'variance' Note 1 C-statistic splitting, C < 0.50, consider statistic value 1 - C allow good 'anti-predictive' splits. , C-statistic initially computed 0.1, considered 1 - 0.10 = 0.90. Note 2 Gini impurity, value 0 1 usually indicate best worst possible scores, respectively. make things simple avoid introducing split_max_stat input, flip values Gini impurity 1 0 indicate best worst possible scores, respectively. oobag_pred_type (character) type --bag predictions compute fitting ensemble. Valid options tree type: 'none' : compute --bag predictions 'leaf' : ID predicted leaf returned tree Valid options survival: 'risk' : probability event occurring oobag_pred_horizon (default). 'surv' : 1 - risk. 'chf' : cumulative hazard function oobag_pred_horizon. 'mort' : mortality, .e., number events expected observations training data identical given observation. Valid options classification: 'prob' : probability class (default) 'class' : class (.e., .max(prob)) Valid options regression: 'mean' : mean value (default) oobag_pred_horizon (numeric) numeric value indicating time used --bag predictions. Default median observed times, .e., oobag_pred_horizon = median(time). input relevant survival trees prediction type 'risk', 'surv', 'chf'. oobag_eval_every (integer) --bag performance ensemble checked every oobag_eval_every trees. , oobag_eval_every = 10, --bag performance checked growing 10th tree, 20th tree, . Default oobag_eval_every = n_tree. oobag_fun (function) used evaluating --bag prediction accuracy every oobag_eval_every trees. oobag_fun = NULL (default), evaluation statistic selected based tree type survival: Harrell's C-statistic (1982) classification: Area underneath ROC curve (AUC-ROC) regression: Traditional prediction R-squared use oobag_fun note following: oobag_fun three inputs: y_mat, w_vec, s_vec survival trees, y_mat two column matrix first column named 'time' second named 'status'. classification trees, y_mat matrix number columns = number distinct classes outcome. regression, y_mat matrix one column. s_vec numeric vector containing predictions oobag_fun return numeric output length 1 details, see --bag vignette. importance (character) Indicate method variable importance: 'none': variable importance computed. 'anova': compute analysis variance (ANOVA) importance 'negate': compute negation importance 'permute': compute permutation importance details methods, see orsf_vi. importance_max_pvalue (double) relevant importance \"anova\". maximum p-value register positive case counting number times variable found 'significant' tree growth. Default 0.01, recommended Menze et al. group_factors (logical) relevant variable importance estimated. TRUE, importance factor variables reported overall aggregating importance individual levels factor. FALSE, importance individual factor levels returned. tree_seeds (integer vector) Optional. specified, random seeds set using values tree_seeds[] growing tree . Two forests grown number trees seeds exact --bag samples, making --bag error estimates forests comparable. NULL (default), seeds picked random. attach_data (logical) TRUE, copy training data attached output. required plan using functions like orsf_pd_oob orsf_summarize_uni interpret forest using training data. Default TRUE. no_fit (logical) TRUE, model fitting steps defined saved, training initiated. object returned can directly submitted orsf_train() long attach_data TRUE. na_action (character) happen data contains missing values (.e., NA values). Valid options : 'fail' : error thrown data contains NA values 'omit' : rows data incomplete data dropped 'impute_meanmode' : missing values continuous categorical variables data imputed using mean mode, respectively. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used). object untrained 'aorsf' object, created setting no_fit = TRUE orsf().","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Oblique Random Forests — orsf","text":"obliqueForest object","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Oblique Random Forests — orsf","text":"function called orf()? earlier versions, aorsf package exclusively oblique random survival forests. formula survival oblique RFs: response formula can survival object returned Surv function, can also just time status variables. .e., Surv(time, status) ~ . works time + status ~ . works response can also survival object stored data. example, y ~ . valid formula data$y inherits Surv class. mtry: mtry parameter may temporarily reduced ensure linear models used find combinations predictors remain stable. occurs coefficients linear model fitting algorithms may become infinite number predictors exceeds number observations. oobag_fun: oobag_fun specified, used compute negation importance permutation importance, role ANOVA importance. n_thread: R function called C++ (.e., user-supplied function compute --bag error identify linear combinations variables), n_thread automatically set 1 attempting run R functions multiple threads cause R session crash.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"what-is-an-oblique-decision-tree-","dir":"Reference","previous_headings":"","what":"What is an oblique decision tree?","title":"Oblique Random Forests — orsf","text":"Decision trees developed splitting set training data two new subsets, goal similarity within new subsets . splitting process repeated resulting subsets data stopping criterion met. new subsets data formed based single predictor, decision tree said axis-based splits data appear perpendicular axis predictor. linear combinations variables used instead single variable, tree oblique splits data neither parallel right angle axis Figure : Decision trees classification axis-based splitting (left) oblique splitting (right). Cases orange squares; controls purple circles. trees partition predictor space defined variables X1 X2, oblique splits better job separating two classes.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"what-is-a-random-forest-","dir":"Reference","previous_headings":"","what":"What is a random forest?","title":"Oblique Random Forests — orsf","text":"Random forests collections de-correlated decision trees. Predictions tree aggregated make ensemble prediction forest. details, see Breiman el, 2001.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"training-out-of-bag-error-and-testing","dir":"Reference","previous_headings":"","what":"Training, out-of-bag error, and testing","title":"Oblique Random Forests — orsf","text":"random forests, tree grown bootstrapped version training set. bootstrap samples selected replacement, bootstrapped training set contains two-thirds instances original training set. '--bag' data instances bootstrapped training set. tree random forest can make predictions --bag data, --bag predictions can aggregated make ensemble --bag prediction. Since --bag data used grow tree, accuracy ensemble --bag predictions approximate generalization error random forest. Generalization error refers error random forest's predictions applied predict outcomes data used train , .e., testing data.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Oblique Random Forests — orsf","text":"orsf() entry-point aorsf package. can used fit classification, regression, survival forests. classification, fit oblique RF predict penguin species using penguin data magnificent palmerpenguins R package regression, use data predict bill length penguins: personal favorite oblique survival RF accelerated Cox regression first type oblique RF aorsf provided (see ArXiv paper; paper also published Journal Computational Graphical Statistics publicly available ). , use predict mortality risk following diagnosis primary biliary cirrhosis:","code":"library(aorsf) library(magrittr) # for %>% # An oblique classification RF penguin_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = species ~ .) penguin_fit ## ---------- Oblique random classification forest ## ## Linear combinations: Accelerated Logistic regression ## N observations: 333 ## N classes: 3 ## N trees: 5 ## N predictors total: 7 ## N predictors per node: 3 ## Average leaves per tree: 4.4 ## Min observations in leaf: 5 ## OOB stat value: 0.99 ## OOB stat type: AUC-ROC ## Variable importance: anova ## ## ----------------------------------------- # An oblique regression RF bill_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = bill_length_mm ~ .) bill_fit ## ---------- Oblique random regression forest ## ## Linear combinations: Accelerated Linear regression ## N observations: 333 ## N trees: 5 ## N predictors total: 7 ## N predictors per node: 3 ## Average leaves per tree: 51.2 ## Min observations in leaf: 5 ## OOB stat value: 0.77 ## OOB stat type: RSQ ## Variable importance: anova ## ## ----------------------------------------- # An oblique survival RF pbc_fit <- orsf(data = pbc_orsf, n_tree = 5, formula = Surv(time, status) ~ . - id) pbc_fit ## ---------- Oblique random survival forest ## ## Linear combinations: Accelerated Cox regression ## N observations: 276 ## N events: 111 ## N trees: 5 ## N predictors total: 17 ## N predictors per node: 5 ## Average leaves per tree: 20 ## Min observations in leaf: 5 ## Min events in leaf: 1 ## OOB stat value: 0.76 ## OOB stat type: Harrell's C-index ## Variable importance: anova ## ## -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"more-than-one-way-to-grow-a-forest","dir":"Reference","previous_headings":"","what":"More than one way to grow a forest","title":"Oblique Random Forests — orsf","text":"can use orsf(no_fit = TRUE) make specification grow forest instead fitted forest. ? Two reasons: computational tasks, may want check long take fit forest commit : fitting multiple forests, use blueprint along orsf_train() orsf_update() simplify code:","code":"orsf_spec <- orsf(pbc_orsf, formula = time + status ~ . - id, no_fit = TRUE) orsf_spec ## Untrained oblique random survival forest ## ## Linear combinations: Accelerated Cox regression ## N observations: 276 ## N events: 111 ## N trees: 500 ## N predictors total: 17 ## N predictors per node: 5 ## Average leaves per tree: 0 ## Min observations in leaf: 5 ## Min events in leaf: 1 ## OOB stat value: none ## OOB stat type: Harrell's C-index ## Variable importance: anova ## ## ----------------------------------------- orsf_spec %>% orsf_update(n_tree = 10000) %>% orsf_time_to_train() ## Time difference of 2.4331 secs orsf_fit <- orsf_train(orsf_spec) orsf_fit_10 <- orsf_update(orsf_fit, leaf_min_obs = 10) orsf_fit_20 <- orsf_update(orsf_fit, leaf_min_obs = 20) orsf_fit$leaf_min_obs ## [1] 5 orsf_fit_10$leaf_min_obs ## [1] 10 orsf_fit_20$leaf_min_obs ## [1] 20"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"tidymodels","dir":"Reference","previous_headings":"","what":"tidymodels","title":"Oblique Random Forests — orsf","text":"tidymodels includes support aorsf computational engine: Prediction aorsf models different times also supported:","code":"library(tidymodels) library(censored) library(yardstick) pbc_tidy <- pbc_orsf %>% mutate(event_time = Surv(time, status), .before = 1) %>% select(-c(id, time, status)) %>% as_tibble() split <- initial_split(pbc_tidy) orsf_spec <- rand_forest() %>% set_engine(\"aorsf\") %>% set_mode(\"censored regression\") orsf_fit <- fit(orsf_spec, formula = event_time ~ ., data = training(split)) time_points <- seq(500, 3000, by = 500) test_pred <- augment(orsf_fit, new_data = testing(split), eval_time = time_points) brier_scores <- test_pred %>% brier_survival(truth = event_time, .pred) brier_scores ## # A tibble: 6 x 4 ## .metric .estimator .eval_time .estimate ## ## 1 brier_survival standard 500 0.0466 ## 2 brier_survival standard 1000 0.0754 ## 3 brier_survival standard 1500 0.0612 ## 4 brier_survival standard 2000 0.0885 ## 5 brier_survival standard 2500 0.133 ## 6 brier_survival standard 3000 0.141 roc_scores <- test_pred %>% roc_auc_survival(truth = event_time, .pred) roc_scores ## # A tibble: 6 x 4 ## .metric .estimator .eval_time .estimate ## ## 1 roc_auc_survival standard 500 0.947 ## 2 roc_auc_survival standard 1000 0.939 ## 3 roc_auc_survival standard 1500 0.982 ## 4 roc_auc_survival standard 2000 0.961 ## 5 roc_auc_survival standard 2500 0.929 ## 6 roc_auc_survival standard 3000 0.942"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Oblique Random Forests — orsf","text":"Harrell, E F, Califf, M R, Pryor, B D, Lee, L K, Rosati, R (1982). \"Evaluating yield medical tests.\" Jama, 247(18), 2543-2546. Breiman, Leo (2001). \"Random Forests.\" Machine Learning, 45(1), 5-32. ISSN 1573-0565. Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS (2008). \"Random survival forests.\" Annals Applied Statistics, 2(3). Menze, H B, Kelm, Michael B, Splitthoff, N D, Koethe, Ullrich, Hamprecht, F (2011). \"oblique random forests.\" Machine Learning Knowledge Discovery Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011, Proceedings, Part II 22, 453-469. Springer. Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min Y, Mcclure LA, Howard G, Simon N (2019). \"Oblique random survival forests.\" Annals Applied Statistics, 13(3). Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM (2023). \"Accelerated interpretable oblique random survival forests.\" Journal Computational Graphical Statistics, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":null,"dir":"Reference","previous_headings":"","what":"Oblique random forest control — orsf_control","title":"Oblique random forest control — orsf_control","text":"Oblique random forest control","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Oblique random forest control — orsf_control","text":"","code":"orsf_control( tree_type, method, scale_x, ties, net_mix, target_df, max_iter, epsilon, ... ) orsf_control_classification( method = \"glm\", scale_x = TRUE, net_mix = 0.5, target_df = NULL, max_iter = 20, epsilon = 1e-09, ... ) orsf_control_regression( method = \"glm\", scale_x = TRUE, net_mix = 0.5, target_df = NULL, max_iter = 20, epsilon = 1e-09, ... ) orsf_control_survival( method = \"glm\", scale_x = TRUE, ties = \"efron\", net_mix = 0.5, target_df = NULL, max_iter = 20, epsilon = 1e-09, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Oblique random forest control — orsf_control","text":"tree_type (character) type tree. Valid options \"classification\", .e., categorical outcomes \"regression\", .e., continuous outcomes \"survival\", .e., time-event outcomes method (character function) identify linear linear combinations predictors. method character value, must one : 'glm': linear, logistic, cox regression 'net': 'glm' penalty terms 'pca': principal component analysis 'random': random draw uniform distribution method function, used identify linear combinations predictor variables. method must case accept three inputs named x_node, y_node w_node, expect following types dimensions: x_node (matrix; n rows, p columns) y_node (matrix; n rows, 2 columns) w_node (matrix; n rows, 1 column) addition, method must return matrix p rows 1 column. scale_x (logical) TRUE, values predictors scaled prior instance finding linear combination predictors, using summary values data current node decision tree. ties (character) character string specifying method tie handling. relevant modeling survival outcomes using method engages tied outcome times. ties, methods equivalent. Valid options 'breslow' 'efron'. Efron approximation default accurate dealing tied event times similar computational efficiency compared Breslow method. net_mix (double) elastic net mixing parameter. value 1 gives lasso penalty, value 0 gives ridge penalty. multiple values alpha given, penalized model fit using alpha value prior splitting node. target_df (integer) Preferred number variables used linear combination. example, mtry 5 target_df 3, sample 5 predictors look best linear combination using 3 . max_iter (integer) iteration continues convergence (see eps ) number attempted iterations equal iter_max. epsilon (double) using modeling based method, iteration continues algorithm relative change kind objective less epsilon, absolute change less sqrt(epsilon). ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Oblique random forest control — orsf_control","text":"object class 'orsf_control', used input control argument orsf. Components : tree_type: type trees fit lincomb_type: method linear combinations lincomb_eps: epsilon convergence lincomb_iter_max: max iterations lincomb_scale: scale . lincomb_alpha: mixing parameter lincomb_df_target: target degrees freedom lincomb_ties_method: method ties survival time lincomb_R_function: R function custom splits","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Oblique random forest control — orsf_control","text":"Adjust scale_x risk. Setting scale_x = FALSE reduce computation time also make orsf model dependent scale data, default value TRUE.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Oblique random forest control — orsf_control","text":"First load relevant packages","code":"set.seed(329730) suppressPackageStartupMessages({ library(aorsf) library(survival) library(ranger) library(riskRegression) })"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"accelerated-linear-combinations","dir":"Reference","previous_headings":"","what":"Accelerated linear combinations","title":"Oblique random forest control — orsf_control","text":"accelerated ORSF ensemble default nice balance computational speed prediction accuracy. runs single iteration Newton Raphson scoring Cox partial likelihood function find linear combinations predictors.","code":"fit_accel <- orsf(pbc_orsf, control = orsf_control_survival(), formula = Surv(time, status) ~ . - id, tree_seeds = 329)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"linear-combinations-with-cox-regression","dir":"Reference","previous_headings":"","what":"Linear combinations with Cox regression","title":"Oblique random forest control — orsf_control","text":"Setting inputs orsf_control_survival scale X matrix repeat iterations convergence allows run Cox regression non-terminal node survival tree, using regression coefficients create linear combinations predictors:","code":"control_cph <- orsf_control_survival(method = 'glm', scale_x = TRUE, max_iter = 20) fit_cph <- orsf(pbc_orsf, control = control_cph, formula = Surv(time, status) ~ . - id, tree_seeds = 329)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"linear-combinations-with-penalized-cox-regression","dir":"Reference","previous_headings":"","what":"Linear combinations with penalized cox regression","title":"Oblique random forest control — orsf_control","text":"Setting method == 'net' runs penalized Cox regression non-terminal node survival tree. can really helpful want feature selection within node, lot slower 'glm' option.","code":"# select 3 predictors out of 5 to be used in # each linear combination of predictors. control_net <- orsf_control_survival(method = 'net', target_df = 3) fit_net <- orsf(pbc_orsf, control = control_net, formula = Surv(time, status) ~ . - id, tree_seeds = 329)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"linear-combinations-with-your-own-function","dir":"Reference","previous_headings":"","what":"Linear combinations with your own function","title":"Oblique random forest control — orsf_control","text":"addition built-methods, customized functions can used identify linear combinations predictors. ’ll demonstrate . first uses random coefficients second derives coefficients principal component analysis third uses ranger() inside orsf(). approach similar method known reinforcement learning trees (see RLT package), although method “muting” crude compared method proposed Zhu et al.  can plug functions orsf_control_custom(), pass result orsf(): fit seems work best example? Let’s find evaluating --bag survival predictions. AUC values, highest lowest: indices prediction accuracy: inspection, net, accel, rlt high discrimination index prediction accuracy. rando pca less well, aren’t bad.","code":"f_rando <- function(x_node, y_node, w_node){ matrix(runif(ncol(x_node)), ncol=1) } f_pca <- function(x_node, y_node, w_node) { # estimate two principal components. pca <- stats::prcomp(x_node, rank. = 2) # use the second principal component to split the node pca$rotation[, 1L, drop = FALSE] } f_rlt <- function(x_node, y_node, w_node){ colnames(y_node) <- c('time', 'status') colnames(x_node) <- paste(\"x\", seq(ncol(x_node)), sep = '') data <- as.data.frame(cbind(y_node, x_node)) if(nrow(data) <= 10) return(matrix(runif(ncol(x_node)), ncol = 1)) fit <- ranger::ranger(data = data, formula = Surv(time, status) ~ ., num.trees = 25, num.threads = 1, min.node.size = 5, importance = 'permutation') out <- sort(fit$variable.importance, decreasing = TRUE) # \"mute\" the least two important variables n_vars <- length(out) if(n_vars > 4){ out[c(n_vars, n_vars-1)] <- 0 } # ensure out has same variable order as input out <- out[colnames(x_node)] # protect yourself out[is.na(out)] <- 0 matrix(out, ncol = 1) } fit_rando <- orsf(pbc_orsf, Surv(time, status) ~ . - id, control = orsf_control_survival(method = f_rando), tree_seeds = 329) fit_pca <- orsf(pbc_orsf, Surv(time, status) ~ . - id, control = orsf_control_survival(method = f_pca), tree_seeds = 329) fit_rlt <- orsf(pbc_orsf, time + status ~ . - id, control = orsf_control_survival(method = f_rlt), tree_seeds = 329) risk_preds <- list( accel = fit_accel$pred_oobag, cph = fit_cph$pred_oobag, net = fit_net$pred_oobag, rando = fit_rando$pred_oobag, pca = fit_pca$pred_oobag, rlt = fit_rlt$pred_oobag ) sc <- Score(object = risk_preds, formula = Surv(time, status) ~ 1, data = pbc_orsf, summary = 'IPA', times = fit_accel$pred_horizon) sc$AUC$score[order(-AUC)] ## model times AUC se lower upper ## 1: net 1788 0.9151649 0.02025057 0.8754745 0.9548553 ## 2: rlt 1788 0.9136390 0.02013475 0.8741756 0.9531024 ## 3: accel 1788 0.9095628 0.02143250 0.8675558 0.9515697 ## 4: cph 1788 0.9095628 0.02143250 0.8675558 0.9515697 ## 5: rando 1788 0.9062197 0.02148854 0.8641029 0.9483365 ## 6: pca 1788 0.8999479 0.02226683 0.8563057 0.9435901 sc$Brier$score[order(-IPA), .(model, times, IPA)] ## model times IPA ## 1: net 1788 0.4905777 ## 2: accel 1788 0.4806065 ## 3: cph 1788 0.4806065 ## 4: rlt 1788 0.4687322 ## 5: pca 1788 0.4383995 ## 6: rando 1788 0.4302814 ## 7: Null model 1788 0.0000000"},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":null,"dir":"Reference","previous_headings":"","what":"Cox regression ORSF control — orsf_control_cph","title":"Cox regression ORSF control — orsf_control_cph","text":"Use coefficients proportional hazards model create linear combinations predictor variables fitting orsf model.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Cox regression ORSF control — orsf_control_cph","text":"","code":"orsf_control_cph(method = \"efron\", eps = 1e-09, iter_max = 20, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Cox regression ORSF control — orsf_control_cph","text":"method (character) character string specifying method tie handling. ties, methods equivalent. Valid options 'breslow' 'efron'. Efron approximation default accurate dealing tied event times similar computational efficiency compared Breslow method. eps (double) using Newton Raphson scoring identify linear combinations inputs, iteration continues algorithm relative change log partial likelihood less eps, absolute change less sqrt(eps). Must positive. default value 1e-09 used consistency survival::coxph.control. iter_max (integer) iteration continues convergence (see eps ) number attempted iterations equal iter_max. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Cox regression ORSF control — orsf_control_cph","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Cox regression ORSF control — orsf_control_cph","text":"code survival package modified make routine. details Cox proportional hazards model, see coxph /Therneau Grambsch (2000).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Cox regression ORSF control — orsf_control_cph","text":"Therneau T.M., Grambsch P.M. (2000) Cox Model. : Modeling Survival Data: Extending Cox Model. Statistics Biology Health. Springer, New York, NY. DOI: 10.1007/978-1-4757-3294-8_3","code":""},{"path":[]},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_custom.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Custom ORSF control — orsf_control_custom","text":"","code":"orsf_control_custom(beta_fun, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_custom.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Custom ORSF control — orsf_control_custom","text":"beta_fun (function) function define coefficients used linear combinations predictor variables. beta_fun must accept three inputs named x_node, y_node w_node, expect following types dimensions: x_node (matrix; n rows, p columns) y_node (matrix; n rows, 2 columns) w_node (matrix; n rows, 1 column) addition, beta_fun must return matrix p rows 1 column. conditions met, orsf_control_custom() let know. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_custom.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Custom ORSF control — orsf_control_custom","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":null,"dir":"Reference","previous_headings":"","what":"Accelerated ORSF control — orsf_control_fast","title":"Accelerated ORSF control — orsf_control_fast","text":"Fast methods identify linear combinations predictors fitting orsf model.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Accelerated ORSF control — orsf_control_fast","text":"","code":"orsf_control_fast(method = \"efron\", do_scale = TRUE, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Accelerated ORSF control — orsf_control_fast","text":"method (character) character string specifying method tie handling. ties, methods equivalent. Valid options 'breslow' 'efron'. Efron approximation default accurate dealing tied event times similar computational efficiency compared Breslow method. do_scale (logical) TRUE, values predictors scaled prior instance Newton Raphson scoring, using summary values data current node decision tree. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Accelerated ORSF control — orsf_control_fast","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Accelerated ORSF control — orsf_control_fast","text":"code survival package modified make routine. Adjust do_scale risk. Setting do_scale = FALSE reduce computation time also make orsf model dependent scale data, default value TRUE.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":null,"dir":"Reference","previous_headings":"","what":"Penalized Cox regression ORSF control — orsf_control_net","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"Use regularized Cox proportional hazard models identify linear combinations input variables fitting orsf model.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"","code":"orsf_control_net(alpha = 1/2, df_target = NULL, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"alpha (double) elastic net mixing parameter. value 1 gives lasso penalty, value 0 gives ridge penalty. multiple values alpha given, penalized model fit using alpha value prior splitting node. df_target (integer) Preferred number variables used linear combination. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"df_target less mtry, separate argument orsf indicates number variables chosen random prior finding linear combination variables.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"Simon, Noah, Friedman, Jerome, Hastie, Trevor, Tibshirani, Rob (2011). \"Regularization paths Cox's proportional hazards model via coordinate descent.\" Journal statistical software, 39(5), 1.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":null,"dir":"Reference","previous_headings":"","what":"Individual Conditional Expectations — orsf_ice_oob","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Compute individual conditional expectations oblique random forest. Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor. can compute individual conditional expectations three ways using random forest: using -bag predictions training data using --bag predictions training data using predictions new set data See examples details","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Individual Conditional Expectations — orsf_ice_oob","text":"","code":"orsf_ice_oob( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_ice_inb( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_ice_new( object, pred_spec, new_data, pred_horizon = NULL, pred_type = NULL, na_action = \"fail\", expand_grid = TRUE, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Individual Conditional Expectations — orsf_ice_oob","text":"object (ObliqueForest) trained oblique random forest object (see orsf). pred_spec (named list, pspec_auto, data.frame). pred_spec named list, item list vector values used points partial dependence function. name item list indicate variable modified take corresponding values. pred_spec created using pred_spec_auto(), needed names variables use (see pred_spec_auto). pred_spec data.frame, columns indicate variable names, values indicate variable values, partial dependence computed using inputs row. pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_type (character) type predictions compute. Valid Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class regression: 'mean': predicted mean, .e., expected value expand_grid (logical) TRUE, partial dependence computed possible combinations inputs pred_spec. FALSE, partial dependence computed variable pred_spec, separately. boundary_checks (logical) TRUE, pred_spec checked make sure requested values 10th 90th percentile object's training data. FALSE, checks skipped. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used). new_data data.frame, tibble, data.table compute predictions . na_action (character) happen new_data contains missing values (.e., NA values). Valid options : 'fail' : error thrown new_data contains NA values 'omit' : rows new_data incomplete data dropped","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Individual Conditional Expectations — orsf_ice_oob","text":"data.table containing individual conditional expectations specified variable(s) , relevant, specified prediction horizon(s).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Individual Conditional Expectations — orsf_ice_oob","text":"can compute individual conditional expectation individual conditional expectations three ways: using -bag predictions training data. -bag individual conditional expectation indicates relationships model learned training. helpful goal interpret model. using --bag predictions training data. --bag individual conditional expectation indicates relationships model learned training using --bag data simulates application model new data. helpful want test model’s reliability fairness new data don’t access large testing set. using predictions new set data. New data individual conditional expectation shows model predicts outcomes observations seen. helpful want test model’s reliability fairness.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"classification","dir":"Reference","previous_headings":"","what":"Classification","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Begin fitting oblique classification random forest: Compute individual conditional expectation using --bag data flipper_length_mm = c(190, 210). two identifiers output: id_variable identifier current value variable(s) data. redundant one variable, helpful multiple variables. id_row identifier observation original data. Note predicted probabilities returned class observation data. Predicted probabilities given observation given variable value sum 1. example,","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) ice_oob <- orsf_ice_oob(fit_clsf, pred_spec = pred_spec) ice_oob ## id_variable id_row class flipper_length_mm pred ## 1: 1 1 Adelie 190 0.92059968 ## 2: 1 2 Adelie 190 0.80953569 ## 3: 1 3 Adelie 190 0.84869374 ## 4: 1 4 Adelie 190 0.93559660 ## 5: 1 5 Adelie 190 0.97708693 ## --- ## 896: 2 146 Gentoo 210 0.25636964 ## 897: 2 147 Gentoo 210 0.04798334 ## 898: 2 148 Gentoo 210 0.07945140 ## 899: 2 149 Gentoo 210 0.84811899 ## 900: 2 150 Gentoo 210 0.10695367 ice_oob %>% .[flipper_length_mm == 190] %>% .[id_row == 1] %>% .[['pred']] %>% sum() ## [1] 1"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"regression","dir":"Reference","previous_headings":"","what":"Regression","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Begin fitting oblique regression random forest: Compute individual conditional expectation using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row flipper_length_mm pred ## 1: 1 1 190 37.94483 ## 2: 1 2 190 37.61595 ## 3: 1 3 190 37.53681 ## 4: 1 4 190 39.49476 ## 5: 1 5 190 38.95635 ## --- ## 362: 2 179 210 51.80471 ## 363: 2 180 210 47.27183 ## 364: 2 181 210 47.05031 ## 365: 2 182 210 50.39028 ## 366: 2 183 210 48.44774 pred_spec = pred_spec_auto(species, island, body_mass_g) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row species island body_mass_g pred ## 1: 1 1 Adelie Biscoe 3200 37.78339 ## 2: 1 2 Adelie Biscoe 3200 37.73273 ## 3: 1 3 Adelie Biscoe 3200 37.71248 ## 4: 1 4 Adelie Biscoe 3200 40.25782 ## 5: 1 5 Adelie Biscoe 3200 40.04074 ## --- ## 8231: 45 179 Gentoo Torgersen 5300 46.14559 ## 8232: 45 180 Gentoo Torgersen 5300 43.98050 ## 8233: 45 181 Gentoo Torgersen 5300 44.59837 ## 8234: 45 182 Gentoo Torgersen 5300 44.85146 ## 8235: 45 183 Gentoo Torgersen 5300 44.23710 ice_new <- orsf_ice_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row variable value level pred ## 1: 1 1 species NA Adelie 37.74136 ## 2: 1 2 species NA Adelie 37.42367 ## 3: 1 3 species NA Adelie 37.04598 ## 4: 1 4 species NA Adelie 39.89602 ## 5: 1 5 species NA Adelie 39.14848 ## --- ## 2009: 5 179 body_mass_g 5300 51.50196 ## 2010: 5 180 body_mass_g 5300 47.27055 ## 2011: 5 181 body_mass_g 5300 48.34064 ## 2012: 5 182 body_mass_g 5300 48.75828 ## 2013: 5 183 body_mass_g 5300 48.11020 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') ice_new <- orsf_ice_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row species island pred ## 1: 1 1 Adelie Biscoe 38.52327 ## 2: 1 2 Adelie Biscoe 38.32073 ## 3: 1 3 Adelie Biscoe 37.71248 ## 4: 1 4 Adelie Biscoe 41.68380 ## 5: 1 5 Adelie Biscoe 40.91140 ## --- ## 179: 1 179 Adelie Biscoe 43.09493 ## 180: 1 180 Adelie Biscoe 38.79455 ## 181: 1 181 Adelie Biscoe 39.37734 ## 182: 1 182 Adelie Biscoe 40.71952 ## 183: 1 183 Adelie Biscoe 39.34501"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"survival","dir":"Reference","previous_headings":"","what":"Survival","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Begin fitting oblique survival random forest: Compute individual conditional expectation using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get individual conditional expectation value: Multi-prediction horizon ice comes minimal extra computational cost. Use fine grid time values assess whether predictors time-varying effects.","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) ice_train <- orsf_ice_inb(fit_surv, pred_spec = list(bili = 1:5)) ice_train ## id_variable id_row pred_horizon bili pred ## 1: 1 1 1826.25 1 0.1290317 ## 2: 1 2 1826.25 1 0.1242352 ## 3: 1 3 1826.25 1 0.0963452 ## 4: 1 4 1826.25 1 0.1172367 ## 5: 1 5 1826.25 1 0.2030256 ## --- ## 746: 5 146 1826.25 5 0.7868537 ## 747: 5 147 1826.25 5 0.2012954 ## 748: 5 148 1826.25 5 0.4893605 ## 749: 5 149 1826.25 5 0.4698220 ## 750: 5 150 1826.25 5 0.9557285 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili)) ice_train ## id_variable id_row pred_horizon bili pred ## 1: 1 1 1826.25 0.59 0.11706741 ## 2: 1 2 1826.25 0.59 0.11562173 ## 3: 1 3 1826.25 0.59 0.09110739 ## 4: 1 4 1826.25 0.59 0.10069721 ## 5: 1 5 1826.25 0.59 0.18769751 ## --- ## 746: 5 146 1826.25 7.21 0.82600898 ## 747: 5 147 1826.25 7.21 0.29156437 ## 748: 5 148 1826.25 7.21 0.58220919 ## 749: 5 149 1826.25 7.21 0.54168688 ## 750: 5 150 1826.25 7.21 0.96204106 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) ice_train ## id_variable id_row pred_horizon bili pred ## 1: 1 1 500 0.59 0.008276627 ## 2: 1 1 1000 0.59 0.055715858 ## 3: 1 1 1500 0.59 0.084987224 ## 4: 1 1 2000 0.59 0.123090885 ## 5: 1 1 2500 0.59 0.165214938 ## --- ## 4496: 5 150 1000 7.21 0.835895969 ## 4497: 5 150 1500 7.21 0.932657591 ## 4498: 5 150 2000 7.21 0.965944498 ## 4499: 5 150 2500 7.21 0.970325309 ## 4500: 5 150 3000 7.21 0.979051377"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":null,"dir":"Reference","previous_headings":"","what":"Partial dependence — orsf_pd_oob","title":"Partial dependence — orsf_pd_oob","text":"Compute partial dependence oblique random forest. Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model's prediction. can compute partial dependence three ways using random forest: using -bag predictions training data using --bag predictions training data using predictions new set data See examples details","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Partial dependence — orsf_pd_oob","text":"","code":"orsf_pd_oob( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, prob_values = c(0.025, 0.5, 0.975), prob_labels = c(\"lwr\", \"medn\", \"upr\"), boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_pd_inb( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, prob_values = c(0.025, 0.5, 0.975), prob_labels = c(\"lwr\", \"medn\", \"upr\"), boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_pd_new( object, pred_spec, new_data, pred_horizon = NULL, pred_type = NULL, na_action = \"fail\", expand_grid = TRUE, prob_values = c(0.025, 0.5, 0.975), prob_labels = c(\"lwr\", \"medn\", \"upr\"), boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Partial dependence — orsf_pd_oob","text":"object (ObliqueForest) trained oblique random forest object (see orsf). pred_spec (named list, pspec_auto, data.frame). pred_spec named list, item list vector values used points partial dependence function. name item list indicate variable modified take corresponding values. pred_spec created using pred_spec_auto(), needed names variables use (see pred_spec_auto). pred_spec data.frame, columns indicate variable names, values indicate variable values, partial dependence computed using inputs row. pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_type (character) type predictions compute. Valid Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class regression: 'mean': predicted mean, .e., expected value expand_grid (logical) TRUE, partial dependence computed possible combinations inputs pred_spec. FALSE, partial dependence computed variable pred_spec, separately. prob_values (numeric) vector values 0 1, indicating quantiles used summarize partial dependence values set inputs. prob_values length prob_labels. quantiles calculated based predictions object set values indicated pred_spec. prob_labels (character) vector labels length prob_values, label indicating corresponding value prob_values labelled summarized outputs. prob_labels length prob_values. boundary_checks (logical) TRUE, pred_spec checked make sure requested values 10th 90th percentile object's training data. FALSE, checks skipped. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used). new_data data.frame, tibble, data.table compute predictions . na_action (character) happen new_data contains missing values (.e., NA values). Valid options : 'fail' : error thrown new_data contains NA values 'omit' : rows new_data incomplete data dropped","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Partial dependence — orsf_pd_oob","text":"data.table containing partial dependence values specified variable(s) , relevant, specified prediction horizon(s).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Partial dependence — orsf_pd_oob","text":"Partial dependence number known limitations assumptions users aware (see Hooker, 2021). particular, partial dependence less intuitive >2 predictors examined jointly, assumed feature(s) partial dependence computed correlated features (likely true many cases). Accumulated local effect plots can used (see ) case feature independence valid assumption.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Partial dependence — orsf_pd_oob","text":"can compute partial dependence individual conditional expectations three ways: using -bag predictions training data. -bag partial dependence indicates relationships model learned training. helpful goal interpret model. using --bag predictions training data. --bag partial dependence indicates relationships model learned training using --bag data simulates application model new data. helpful want test model’s reliability fairness new data don’t access large testing set. using predictions new set data. New data partial dependence shows model predicts outcomes observations seen. helpful want test model’s reliability fairness.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"classification","dir":"Reference","previous_headings":"","what":"Classification","title":"Partial dependence — orsf_pd_oob","text":"Begin fitting oblique classification random forest: Compute partial dependence using --bag data flipper_length_mm = c(190, 210). Note predicted probabilities returned class probabilities mean column sum 1 take sum class specific value pred_spec variables. example, isn’t case median predicted probability!","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_oob <- orsf_pd_oob(fit_clsf, pred_spec = pred_spec) pd_oob ## class flipper_length_mm mean lwr medn upr ## 1: Adelie 190 0.6180632 0.207463688 0.76047056 0.9809703 ## 2: Adelie 210 0.4346177 0.018583256 0.56486883 0.8647387 ## 3: Chinstrap 190 0.2119948 0.017692341 0.15658268 0.7163635 ## 4: Chinstrap 210 0.1801186 0.020454479 0.09525310 0.7085293 ## 5: Gentoo 190 0.1699420 0.001277844 0.02831331 0.5738689 ## 6: Gentoo 210 0.3852637 0.068685035 0.20853993 0.9537020 sum(pd_oob[flipper_length_mm == 190, mean]) ## [1] 1 sum(pd_oob[flipper_length_mm == 190, medn]) ## [1] 0.9453666"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"regression","dir":"Reference","previous_headings":"","what":"Regression","title":"Partial dependence — orsf_pd_oob","text":"Begin fitting oblique regression random forest: Compute partial dependence using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new ## flipper_length_mm mean lwr medn upr ## 1: 190 42.96571 37.09805 43.69769 48.72301 ## 2: 210 45.66012 40.50693 46.31577 51.65163 pred_spec = pred_spec_auto(species, island, body_mass_g) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new ## species island body_mass_g mean lwr medn upr ## 1: Adelie Biscoe 3200 40.31374 37.24373 40.31967 44.22824 ## 2: Chinstrap Biscoe 3200 45.10582 42.63342 45.10859 47.60119 ## 3: Gentoo Biscoe 3200 42.81649 40.19221 42.55664 46.84035 ## 4: Adelie Dream 3200 40.16219 36.95895 40.34633 43.90681 ## 5: Chinstrap Dream 3200 46.21778 43.53954 45.90929 49.19173 ## 6: Gentoo Dream 3200 42.60465 39.89647 42.63520 46.28769 ## 7: Adelie Torgersen 3200 39.91652 36.80227 39.79806 43.68842 ## 8: Chinstrap Torgersen 3200 44.27807 41.95470 44.40742 46.68848 ## 9: Gentoo Torgersen 3200 42.09510 39.49863 41.80049 45.81833 ## 10: Adelie Biscoe 3550 40.77971 38.04027 40.59561 44.57505 ## 11: Chinstrap Biscoe 3550 45.81304 43.52102 45.73116 48.36366 ## 12: Gentoo Biscoe 3550 43.31233 40.77355 43.03077 47.22936 ## 13: Adelie Dream 3550 40.77741 38.07399 40.78175 44.37273 ## 14: Chinstrap Dream 3550 47.30926 44.80493 46.77540 50.47092 ## 15: Gentoo Dream 3550 43.26955 40.86119 43.16204 46.89190 ## 16: Adelie Torgersen 3550 40.25780 37.35251 40.07871 44.04576 ## 17: Chinstrap Torgersen 3550 44.77911 42.60161 44.81944 47.14986 ## 18: Gentoo Torgersen 3550 42.49520 39.95866 42.14160 46.26237 ## 19: Adelie Biscoe 3975 41.61744 38.94515 41.36634 45.38752 ## 20: Chinstrap Biscoe 3975 46.59363 44.59970 46.44923 49.11457 ## 21: Gentoo Biscoe 3975 44.07857 41.60792 43.74562 47.85109 ## 22: Adelie Dream 3975 41.50511 39.06187 41.24741 45.13027 ## 23: Chinstrap Dream 3975 48.14978 45.87390 47.54867 51.50683 ## 24: Gentoo Dream 3975 44.01928 41.70577 43.84099 47.50470 ## 25: Adelie Torgersen 3975 40.94764 38.12519 40.66759 44.73689 ## 26: Chinstrap Torgersen 3975 45.44820 43.49986 45.44036 47.63243 ## 27: Gentoo Torgersen 3975 43.13791 40.70628 42.70627 46.87306 ## 28: Adelie Biscoe 4700 42.93914 40.48463 42.44768 46.81756 ## 29: Chinstrap Biscoe 4700 47.18534 45.40866 47.07739 49.55747 ## 30: Gentoo Biscoe 4700 45.32541 43.08173 44.93498 49.23391 ## 31: Adelie Dream 4700 42.73806 40.44229 42.22226 46.49936 ## 32: Chinstrap Dream 4700 48.37354 46.34335 48.00781 51.18955 ## 33: Gentoo Dream 4700 45.09132 42.88328 44.79530 48.82180 ## 34: Adelie Torgersen 4700 42.09349 39.72074 41.56168 45.68838 ## 35: Chinstrap Torgersen 4700 46.17045 44.39042 46.09525 48.35127 ## 36: Gentoo Torgersen 4700 44.31621 42.18968 43.81773 47.98024 ## 37: Adelie Biscoe 5300 43.89769 41.43335 43.28504 48.10892 ## 38: Chinstrap Biscoe 5300 47.53721 45.66038 47.52770 49.88701 ## 39: Gentoo Biscoe 5300 46.16115 43.81722 45.59309 50.57469 ## 40: Adelie Dream 5300 43.59846 41.25825 43.24518 47.46193 ## 41: Chinstrap Dream 5300 48.48139 46.36282 48.25679 51.02996 ## 42: Gentoo Dream 5300 45.91819 43.62832 45.54110 49.91622 ## 43: Adelie Torgersen 5300 42.92879 40.66576 42.31072 46.76406 ## 44: Chinstrap Torgersen 5300 46.59576 44.80400 46.49196 49.03906 ## 45: Gentoo Torgersen 5300 45.11384 42.95190 44.51289 49.27629 ## species island body_mass_g mean lwr medn upr pd_new <- orsf_pd_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new ## variable value level mean lwr medn upr ## 1: species NA Adelie 41.90271 37.10417 41.51723 48.51478 ## 2: species NA Chinstrap 47.11314 42.40419 46.96478 51.51392 ## 3: species NA Gentoo 44.37038 39.87306 43.89889 51.21635 ## 4: island NA Biscoe 44.21332 37.22711 45.27862 51.21635 ## 5: island NA Dream 44.43354 37.01471 45.57261 51.51392 ## 6: island NA Torgersen 43.29539 37.01513 44.26924 49.84391 ## 7: body_mass_g 3200 42.84625 37.03978 43.95991 49.19173 ## 8: body_mass_g 3550 43.53326 37.56730 44.43756 50.47092 ## 9: body_mass_g 3975 44.30431 38.31567 45.22089 51.50683 ## 10: body_mass_g 4700 45.22559 39.88199 46.34680 51.18955 ## 11: body_mass_g 5300 45.91412 40.84742 46.95327 51.48851 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') pd_new <- orsf_pd_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) pd_new ## species island mean lwr medn upr ## 1: Adelie Biscoe 41.98024 37.22711 41.65252 48.51478"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"survival","dir":"Reference","previous_headings":"","what":"Survival","title":"Partial dependence — orsf_pd_oob","text":"Begin fitting oblique survival random forest: Compute partial dependence using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get partial dependence value: vector-valued pred_horizon input comes minimal extra computational cost. Use fine grid time values assess whether predictors time-varying effects. (see partial dependence vignette example)","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) pd_train <- orsf_pd_inb(fit_surv, pred_spec = list(bili = 1:5)) pd_train ## pred_horizon bili mean lwr medn upr ## 1: 1826.25 1 0.2566200 0.02234786 0.1334170 0.8918909 ## 2: 1826.25 2 0.3121392 0.06853733 0.1896849 0.9204338 ## 3: 1826.25 3 0.3703242 0.11409793 0.2578505 0.9416791 ## 4: 1826.25 4 0.4240692 0.15645214 0.3331057 0.9591581 ## 5: 1826.25 5 0.4663670 0.20123406 0.3841700 0.9655296 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili)) pd_train ## pred_horizon bili mean lwr medn upr ## 1: 1826.25 0.590 0.2484695 0.02035041 0.1243120 0.8823385 ## 2: 1826.25 0.725 0.2508045 0.02060111 0.1274237 0.8836536 ## 3: 1826.25 1.500 0.2797763 0.03964900 0.1601715 0.9041584 ## 4: 1826.25 3.500 0.3959349 0.13431288 0.2920400 0.9501230 ## 5: 1826.25 7.210 0.5344511 0.27869513 0.4651185 0.9782084 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) pd_train ## pred_horizon bili mean lwr medn upr ## 1: 500 0.590 0.06184375 0.0004433990 0.008765301 0.5918852 ## 2: 1000 0.590 0.14210619 0.0057937418 0.056124198 0.7381107 ## 3: 1500 0.590 0.20859307 0.0136094784 0.091808079 0.8577223 ## 4: 2000 0.590 0.26823465 0.0230476894 0.145707217 0.8918696 ## 5: 2500 0.590 0.31809404 0.0631155452 0.202189830 0.9035026 ## 6: 3000 0.590 0.39152139 0.0911566314 0.302738552 0.9239861 ## 7: 500 0.725 0.06255088 0.0004462367 0.008934806 0.5980510 ## 8: 1000 0.725 0.14337233 0.0063321712 0.056348007 0.7447805 ## 9: 1500 0.725 0.21058059 0.0140736894 0.093113771 0.8597396 ## 10: 2000 0.725 0.27056356 0.0235448705 0.146307939 0.8941464 ## 11: 2500 0.725 0.31922691 0.0626303822 0.202462648 0.9073970 ## 12: 3000 0.725 0.39426313 0.0911457406 0.308440546 0.9252028 ## 13: 500 1.500 0.06679162 0.0012717884 0.011028398 0.6241228 ## 14: 1000 1.500 0.15727919 0.0114789623 0.068332010 0.7678732 ## 15: 1500 1.500 0.23316655 0.0287320952 0.117289745 0.8789647 ## 16: 2000 1.500 0.30139227 0.0467927208 0.180096425 0.9144202 ## 17: 2500 1.500 0.35260943 0.0845866747 0.238015966 0.9266065 ## 18: 3000 1.500 0.43512074 0.1311103304 0.346025144 0.9438562 ## 19: 500 3.500 0.08638646 0.0052087533 0.028239001 0.6740930 ## 20: 1000 3.500 0.22353655 0.0519179775 0.139604845 0.8283986 ## 21: 1500 3.500 0.32700976 0.0901983241 0.217982772 0.9371150 ## 22: 2000 3.500 0.41618105 0.1445328597 0.311508093 0.9566091 ## 23: 2500 3.500 0.49248461 0.2195110942 0.402095677 0.9636221 ## 24: 3000 3.500 0.56008108 0.2635698957 0.503253258 0.9734948 ## 25: 500 7.210 0.12550962 0.0220920570 0.063425987 0.7526581 ## 26: 1000 7.210 0.32567558 0.1353851175 0.259047345 0.8875150 ## 27: 1500 7.210 0.46327019 0.2181840827 0.386681920 0.9700903 ## 28: 2000 7.210 0.55042753 0.2912654769 0.483477295 0.9812223 ## 29: 2500 7.210 0.61937483 0.3709845684 0.567895754 0.9844945 ## 30: 3000 7.210 0.67963922 0.4247511750 0.645083041 0.9888637 ## pred_horizon bili mean lwr medn upr"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Partial dependence — orsf_pd_oob","text":"Hooker, Giles, Mentch, Lucas, Zhou, Siyu (2021). \"Unrestricted permutation forces extrapolation: variable importance requires least one model, free variable importance.\" Statistics Computing, 31, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":null,"dir":"Reference","previous_headings":"","what":"Scale input data — orsf_scale_cph","title":"Scale input data — orsf_scale_cph","text":"functions exported users may access internal routines used scale inputs orsf_control_cph used.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Scale input data — orsf_scale_cph","text":"","code":"orsf_scale_cph(x_mat, w_vec = NULL) orsf_unscale_cph(x_mat)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Scale input data — orsf_scale_cph","text":"x_mat (numeric matrix) matrix values scaled unscaled. Note orsf_unscale_cph accept x_mat inputs attribute containing transform values, added automatically orsf_scale_cph. w_vec (numeric vector) optional vector weights. weights supplied (default), observations equally weighted. supplied, w_vec must length equal nrow(x_mat).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Scale input data — orsf_scale_cph","text":"scaled unscaled x_mat.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Scale input data — orsf_scale_cph","text":"data transformed first subtracting mean multiplying scale. inverse transform can completed using orsf_unscale_cph dividing column corresponding scale adding mean. values means scales stored attribute output returned orsf_scale_cph (see examples)","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Scale input data — orsf_scale_cph","text":"","code":"x_mat <- as.matrix(pbc_orsf[, c('bili', 'age', 'protime')]) head(x_mat) #> bili age protime #> 1 14.5 58.76523 12.2 #> 2 1.1 56.44627 10.6 #> 3 1.4 70.07255 12.0 #> 4 1.8 54.74059 10.3 #> 5 3.4 38.10541 10.9 #> 7 1.0 55.53457 9.7 x_scaled <- orsf_scale_cph(x_mat) head(x_scaled) #> bili age protime #> [1,] 3.77308887 1.0412574 1.9694656 #> [2,] -0.75476469 0.7719344 -0.1822316 #> [3,] -0.65339483 2.3544852 1.7005035 #> [4,] -0.51823502 0.5738373 -0.5856748 #> [5,] 0.02240421 -1.3581657 0.2212116 #> [6,] -0.78855464 0.6660494 -1.3925613 attributes(x_scaled) # note the transforms attribute #> $dim #> [1] 276 3 #> #> $dimnames #> $dimnames[[1]] #> NULL #> #> $dimnames[[2]] #> [1] \"bili\" \"age\" \"protime\" #> #> #> $transforms #> mean scale #> [1,] 3.333696 0.3378995 #> [2,] 49.799661 0.1161396 #> [3,] 10.735507 1.3448108 #> x_unscaled <- orsf_unscale_cph(x_scaled) head(x_unscaled) #> bili age protime #> [1,] 14.5 58.76523 12.2 #> [2,] 1.1 56.44627 10.6 #> [3,] 1.4 70.07255 12.0 #> [4,] 1.8 54.74059 10.3 #> [5,] 3.4 38.10541 10.9 #> [6,] 1.0 55.53457 9.7 # numeric difference in x_mat and x_unscaled should be practically 0 max(abs(x_mat - x_unscaled)) #> [1] 3.552714e-15"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":null,"dir":"Reference","previous_headings":"","what":"Univariate summary — orsf_summarize_uni","title":"Univariate summary — orsf_summarize_uni","text":"Summarize univariate information ORSF object","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Univariate summary — orsf_summarize_uni","text":"","code":"orsf_summarize_uni( object, n_variables = NULL, pred_horizon = NULL, pred_type = NULL, importance = NULL, verbose_progress = FALSE, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Univariate summary — orsf_summarize_uni","text":"object (ObliqueForest) trained oblique random forest object (see orsf). n_variables (integer) many variables summarized? Setting input lower number reduce computation time. pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_type (character) type predictions compute. Valid Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class regression: 'mean': predicted mean, .e., expected value importance (character) Indicate method variable importance: 'none': variable importance computed. 'anova': compute analysis variance (ANOVA) importance 'negate': compute negation importance 'permute': compute permutation importance details methods, see orsf_vi. verbose_progress (logical) TRUE, progress printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Univariate summary — orsf_summarize_uni","text":"object class 'orsf_summary', includes data importance individual predictors. expected values predictions specific values predictors.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Univariate summary — orsf_summarize_uni","text":"pred_horizon left unspecified, median value time--event variable object's training data used. recommended always specify prediction horizon, median time may especially meaningful horizon compute predicted risk values . object already variable importance values, can safely bypass computation variable importance function setting importance = 'none'.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Univariate summary — orsf_summarize_uni","text":"","code":"object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 25) # since anova importance was used to make object, it is also # used for ranking variables in the summary, unless we specify # a different type of importance orsf_summarize_uni(object, n_variables = 2) #> #> -- ascites (VI Rank: 1) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3112269 0.2184047 0.04443568 0.535574 #> 1 0.4763575 0.3918262 0.24704845 0.714196 #> #> -- edema (VI Rank: 2) -------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3054237 0.1884637 0.04443568 0.5342492 #> 0.5 0.3531583 0.2543017 0.05655187 0.5969800 #> 1 0.4559456 0.4038856 0.23603721 0.6813184 #> #> Predicted risk at time t = 1788 for top 2 predictors # if we want to summarize object according to variables # ranked by negation importance, we can compute negation # importance within orsf_summarize_uni() as follows: orsf_summarize_uni(object, n_variables = 2, importance = 'negate') #> #> -- bili (VI Rank: 1) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2224069 0.1254764 0.03301986 0.3207543 #> 0.80 0.2262946 0.1254764 0.03266947 0.3316953 #> 1.40 0.2536990 0.1646659 0.05763091 0.3902082 #> 3.52 0.3911796 0.3181171 0.17745485 0.6078190 #> 7.25 0.5031346 0.4685324 0.33257259 0.6883006 #> #> -- copper (VI Rank: 2) ------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 25.5 0.2697228 0.1414728 0.03983757 0.4498140 #> 42.8 0.2709936 0.1391208 0.03839266 0.4498140 #> 74.0 0.2865622 0.1740328 0.04594795 0.4755196 #> 129 0.3404841 0.2498839 0.09485429 0.5258744 #> 214 0.4435109 0.3924181 0.20542624 0.6589764 #> #> Predicted risk at time t = 1788 for top 2 predictors"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":null,"dir":"Reference","previous_headings":"","what":"Estimate training time — orsf_time_to_train","title":"Estimate training time — orsf_time_to_train","text":"Estimate training time","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Estimate training time — orsf_time_to_train","text":"","code":"orsf_time_to_train(object, n_tree_subset = NULL)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Estimate training time — orsf_time_to_train","text":"object untrained aorsf object n_tree_subset (integer) many trees fit order estimate time needed train object. default value 10% trees specified object. .e., object n_tree 500, default value n_tree_subset 50.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Estimate training time — orsf_time_to_train","text":"difftime object.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Estimate training time — orsf_time_to_train","text":"","code":"# specify but do not train the model by setting no_fit = TRUE. object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 10, no_fit = TRUE) # approximate the time it will take to grow 10 trees time_estimated <- orsf_time_to_train(object, n_tree_subset=1) print(time_estimated) #> Time difference of 0.04493713 secs # let's see how close the approximation was time_true_start <- Sys.time() orsf_train(object) time_true_stop <- Sys.time() time_true <- time_true_stop - time_true_start print(time_true) #> Time difference of 0.02969646 secs # error abs(time_true - time_estimated) #> Time difference of 0.01524067 secs"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":null,"dir":"Reference","previous_headings":"","what":"Update Forest Parameters — orsf_update","title":"Update Forest Parameters — orsf_update","text":"Update Forest Parameters","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Update Forest Parameters — orsf_update","text":"","code":"orsf_update(object, ..., modify_in_place = FALSE, no_fit = NULL)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Update Forest Parameters — orsf_update","text":"object (ObliqueForest) oblique random forest object (see orsf). ... arguments plug orsf used define update. arguments include: data formula control weights n_tree n_split n_retry n_thread mtry sample_with_replacement sample_fraction leaf_min_events leaf_min_obs split_rule split_min_events split_min_obs split_min_stat pred_type oobag_pred_horizon oobag_eval_every oobag_fun importance importance_max_pvalue group_factors tree_seeds na_action verbose_progress Note can update control, change type forest. example, go classification regression orsf_update. modify_in_place (logical) TRUE, object modified inputs specified .... cautious, modification place overwrite existing data. FALSE (default), object copied modifications applied copy, leaving original object unmodified. no_fit (logical) TRUE, model fitting steps defined saved, training initiated. object returned can directly submitted orsf_train() long attach_data TRUE.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Update Forest Parameters — orsf_update","text":"ObliqueForest object.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Update Forest Parameters — orsf_update","text":"several dynamic inputs orsf default values NULL. Specifically, inputs control, weights, mtry, split_rule, split_min_stat, pred_type, pred_horizon, oobag_eval_function, tree_seeds, oobag_eval_every. explicit value given inputs call, re-formed. example, initial forest includes 17 predictors, default mtry smallest integer greater equal square root 17, .e., 5. , make updated forest 1 less predictor explicitly say mtry = 5, mtry re-initialized update based available 16 predictors, resulting value mtry 4. done avoid many potential errors occur dynamic outputs re-initialized.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Update Forest Parameters — orsf_update","text":"","code":"if (FALSE) { # initial fit has mtry of 5 fit <- orsf(pbc_orsf, time + status ~ . -id) # note that mtry is now 4 (see details) fit_new <- orsf_update(fit, formula = . ~ . - edema, n_tree = 100) # prevent dynamic updates by specifying inputs you want to freeze. fit_newer <- orsf_update(fit_new, mtry = 2) }"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":null,"dir":"Reference","previous_headings":"","what":"Variable Importance — orsf_vi","title":"Variable Importance — orsf_vi","text":"Estimate importance individual predictor variables using oblique random forests.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Variable Importance — orsf_vi","text":"","code":"orsf_vi( object, group_factors = TRUE, importance = NULL, oobag_fun = NULL, n_thread = NULL, verbose_progress = NULL, ... ) orsf_vi_negate( object, group_factors = TRUE, oobag_fun = NULL, n_thread = NULL, verbose_progress = NULL, ... ) orsf_vi_permute( object, group_factors = TRUE, oobag_fun = NULL, n_thread = NULL, verbose_progress = NULL, ... ) orsf_vi_anova(object, group_factors = TRUE, verbose_progress = NULL, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Variable Importance — orsf_vi","text":"object (ObliqueForest) trained oblique random forest object (see orsf). group_factors (logical) TRUE, importance factor variables reported overall aggregating importance individual levels factor. FALSE, importance individual factor levels returned. importance (character) Indicate method variable importance: 'anova': compute analysis variance (ANOVA) importance 'negate': compute negation importance 'permute': compute permutation importance oobag_fun (function) used evaluating --bag prediction accuracy negating coefficients (importance = 'negate') permuting values predictor (importance = 'permute') oobag_fun = NULL (default), evaluation statistic selected based tree type survival: Harrell's C-statistic (1982) classification: Area underneath ROC curve (AUC-ROC) regression: Traditional prediction R-squared use oobag_fun note following: oobag_fun three inputs: y_mat, w_vec, s_vec survival trees, y_mat two column matrix first column named 'time' second named 'status'. classification trees, y_mat matrix number columns = number distinct classes outcome. regression, y_mat matrix one column. s_vec numeric vector containing predictions oobag_fun return numeric output length 1 oobag_fun used created object initial value --bag prediction accuracy consistent values computed variable importance estimated. details, see --bag vignette. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Variable Importance — orsf_vi","text":"orsf_vi functions return named numeric vector. Names vector predictor variables used object Values vector estimated importance given predictor. returned vector sorted highest lowest value, higher values indicating higher importance.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Variable Importance — orsf_vi","text":"ObliqueForest object grown importance = 'anova', 'negate', 'permute', output vector importance values based requested type importance. However, orsf_vi() can used compute variable importance growing forest compute different type importance. orsf_vi() general purpose function extract compute variable importance estimates ObliqueForest object (see orsf). orsf_vi_negate(), orsf_vi_permute(), orsf_vi_anova() wrappers orsf_vi(). way functions work depends whether object given already variable importance estimates (see examples).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"variable-importance-methods","dir":"Reference","previous_headings":"","what":"Variable importance methods","title":"Variable Importance — orsf_vi","text":"negation importance: variable assessed separately multiplying variable's coefficients -1 determining much model's performance changes. worse model's performance negating coefficients given variable, important variable. technique promising b/c require permutation emphasizes variables larger coefficients linear combinations, also relatively new studied much permutation importance. See Jaeger, (2023) details technique. permutation importance: variable assessed separately randomly permuting variable's values determining much model's performance changes. worse model's performance permuting values given variable, important variable. technique flexible, intuitive, frequently used. also several known limitations analysis variance (ANOVA) importance: p-value computed coefficient linear combination variables decision tree. Importance individual predictor variable proportion times p-value coefficient < 0.01. technique efficient computationally, may effective permutation negation terms selecting signal noise variables. See Menze, 2011 details technique.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"anova-importance","dir":"Reference","previous_headings":"","what":"ANOVA importance","title":"Variable Importance — orsf_vi","text":"default variable importance technique, ANOVA, calculated fit oblique random forest ensemble. ANOVA default fast, may decisive permutation negation techniques variable selection.","code":"fit <- orsf(pbc_orsf, Surv(time, status) ~ . - id) fit ## ---------- Oblique random survival forest ## ## Linear combinations: Accelerated Cox regression ## N observations: 276 ## N events: 111 ## N trees: 500 ## N predictors total: 17 ## N predictors per node: 5 ## Average leaves per tree: 21.022 ## Min observations in leaf: 5 ## Min events in leaf: 1 ## OOB stat value: 0.84 ## OOB stat type: Harrell's C-index ## Variable importance: anova ## ## -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"raw-vi-values","dir":"Reference","previous_headings":"","what":"Raw VI values","title":"Variable Importance — orsf_vi","text":"‘raw’ variable importance values can accessed fit object ‘raw’ values factors aggregated single value. Currently one value k-1 levels k level factor. example, can see edema_1 edema_0.5 importance values edema factor variable levels 0, 0.5, 1.","code":"fit$get_importance_raw() ## [,1] ## trt_placebo 0.06355042 ## age 0.23259259 ## sex_f 0.14700432 ## ascites_1 0.46791708 ## hepato_1 0.14349776 ## spiders_1 0.17371938 ## edema_0.5 0.17459191 ## edema_1 0.51197605 ## bili 0.40590758 ## chol 0.17666667 ## albumin 0.25972156 ## copper 0.28840580 ## alk.phos 0.10614251 ## ast 0.18327491 ## trig 0.12815626 ## platelet 0.09265648 ## protime 0.22656250 ## stage 0.20264766"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"collapse-vi-across-factor-levels","dir":"Reference","previous_headings":"","what":"Collapse VI across factor levels","title":"Variable Importance — orsf_vi","text":"get aggregated values across levels factor, access importance element orsf fit: use orsf_vi() group_factors set TRUE (default) Note can make default returned importance values ungrouped setting group_factors FALSE orsf_vi functions orsf function.","code":"# this assumes you used group_factors = TRUE in orsf() fit$importance ## ascites bili edema copper albumin age protime ## 0.46791708 0.40590758 0.31115216 0.28840580 0.25972156 0.23259259 0.22656250 ## stage ast chol spiders sex hepato trig ## 0.20264766 0.18327491 0.17666667 0.17371938 0.14700432 0.14349776 0.12815626 ## alk.phos platelet trt ## 0.10614251 0.09265648 0.06355042 orsf_vi(fit) ## ascites bili edema copper albumin age protime ## 0.46791708 0.40590758 0.31115216 0.28840580 0.25972156 0.23259259 0.22656250 ## stage ast chol spiders sex hepato trig ## 0.20264766 0.18327491 0.17666667 0.17371938 0.14700432 0.14349776 0.12815626 ## alk.phos platelet trt ## 0.10614251 0.09265648 0.06355042"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"add-vi-to-an-oblique-random-forest","dir":"Reference","previous_headings":"","what":"Add VI to an oblique random forest","title":"Variable Importance — orsf_vi","text":"can fit oblique random forest without VI, add VI later","code":"fit_no_vi <- orsf(pbc_orsf, Surv(time, status) ~ . - id, importance = 'none') # Note: you can't call orsf_vi_anova() on fit_no_vi because anova # VI can only be computed while the forest is being grown. orsf_vi_negate(fit_no_vi) ## bili copper sex protime age stage ## 0.130439814 0.051880867 0.038308025 0.025115249 0.023826061 0.020354822 ## albumin ascites chol ast spiders hepato ## 0.019997729 0.015918292 0.013320469 0.010086726 0.007409116 0.007326714 ## edema trt alk.phos trig platelet ## 0.006844435 0.003214544 0.002517057 0.002469545 0.001056829 orsf_vi_permute(fit_no_vi) ## bili copper age ascites protime ## 0.0592069141 0.0237362075 0.0136479213 0.0130805894 0.0123091354 ## stage albumin chol hepato ast ## 0.0117177661 0.0106414724 0.0064501213 0.0058813969 0.0057753740 ## edema spiders sex trig platelet ## 0.0052171180 0.0048427005 0.0023386947 0.0017883700 0.0013533691 ## alk.phos trt ## 0.0006492029 -0.0009921507"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"oblique-random-forest-and-vi-all-at-once","dir":"Reference","previous_headings":"","what":"Oblique random forest and VI all at once","title":"Variable Importance — orsf_vi","text":"fit oblique random forest compute vi time can still get negation VI fit, needs computed","code":"fit_permute_vi <- orsf(pbc_orsf, Surv(time, status) ~ . - id, importance = 'permute') # get the vi instantly (i.e., it doesn't need to be computed again) orsf_vi_permute(fit_permute_vi) ## bili copper ascites protime albumin ## 0.0571305446 0.0243657146 0.0138318057 0.0133401675 0.0130746154 ## age stage chol ast spiders ## 0.0123610374 0.0102963203 0.0077895394 0.0075250059 0.0048628813 ## edema hepato sex platelet trig ## 0.0046003168 0.0039818730 0.0016891584 0.0012767063 0.0007324402 ## alk.phos trt ## 0.0005128897 -0.0014443967 orsf_vi_negate(fit_permute_vi) ## bili copper sex protime stage age ## 0.123331760 0.052544318 0.037291358 0.024977898 0.023239189 0.021934511 ## albumin ascites chol ast spiders edema ## 0.020586632 0.014229536 0.014053040 0.012227048 0.007643156 0.006832766 ## hepato trt alk.phos trig platelet ## 0.006301693 0.004348705 0.002371797 0.002309396 0.001347035"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"custom-functions-for-vi","dir":"Reference","previous_headings":"","what":"Custom functions for VI","title":"Variable Importance — orsf_vi","text":"default prediction accuracy functions work well time: sometimes want something specific defaults just won’t work. cases, can compute VI function ’d like measure prediction accuracy supplying valid function oobag_fun input. example, use mean absolute error . Higher values considered good aorsf computes prediction accuracy, make function return pseudo R-squared based mean absolute error:","code":"fit_standard <- orsf(penguins_orsf, bill_length_mm ~ ., tree_seeds = 1) # Default method for prediction accuracy with VI is R-squared orsf_vi_permute(fit_standard) ## species flipper_length_mm body_mass_g bill_depth_mm ## 0.3725898166 0.3261834607 0.2225730676 0.1026569498 ## island sex year ## 0.0876071687 0.0844807334 0.0006978493 rsq_mae <- function(y_mat, w_vec, s_vec){ mae_standard <- mean(abs((y_mat - mean(y_mat)) * w_vec)) mae_fit <- mean(abs((y_mat - s_vec) * w_vec)) 1 - mae_fit / mae_standard } fit_custom <- orsf_update(fit_standard, oobag_fun = rsq_mae) # not much changes, but the difference between variables shrinks # and the ordering of sex and island has swapped orsf_vi_permute(fit_custom) ## species flipper_length_mm body_mass_g bill_depth_mm ## 0.206951751 0.193248912 0.140899603 0.076759148 ## sex island year ## 0.073042331 0.050851073 0.003633365"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Variable Importance — orsf_vi","text":"Harrell, E F, Califf, M R, Pryor, B D, Lee, L K, Rosati, R (1982). \"Evaluating yield medical tests.\" Jama, 247(18), 2543-2546. Breiman, Leo (2001). \"Random Forests.\" Machine Learning, 45(1), 5-32. ISSN 1573-0565. Menze, H B, Kelm, Michael B, Splitthoff, N D, Koethe, Ullrich, Hamprecht, F (2011). \"oblique random forests.\" Machine Learning Knowledge Discovery Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011, Proceedings, Part II 22, 453-469. Springer. Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM (2023). \"Accelerated interpretable oblique random survival forests.\" Journal Computational Graphical Statistics, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":null,"dir":"Reference","previous_headings":"","what":"Variable Interactions — orsf_vint","title":"Variable Interactions — orsf_vint","text":"Use variable interaction score described Greenwell et al (2018). method can computationally demanding, using n_thread=0 can substantially reduce time needed compute scores.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Variable Interactions — orsf_vint","text":"","code":"orsf_vint( object, predictors = NULL, n_thread = NULL, verbose_progress = NULL, sep = \"..\" )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Variable Interactions — orsf_vint","text":"object (ObliqueForest) trained oblique random forest object (see orsf) predictors (character) vector length 2 names predictors used object. pairwise interactions predictors scored. NULL (default), predictors used. n_thread (integer) number threads use growing trees, computing predictions, computing importance. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. sep (character) separate names two predictors. default value \"..\" returns names name1..name2","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Variable Interactions — orsf_vint","text":"data.table variable interaction scores","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Variable Interactions — orsf_vint","text":"number possible interactions grows exponentially based number predictors. caution warranted using large predictor sets recommended supply specific vector predictor names assess rather global search. good strategy use n_tree = 5 search predictors, pick top 10 interactions, get unique predictors , re-run just predictors trees.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Variable Interactions — orsf_vint","text":"Greenwell, M B, Boehmke, C B, McCarthy, J (2018). \"simple effective model-based variable importance measure.\" arXiv preprint arXiv:1805.04755.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Variable Interactions — orsf_vint","text":"","code":"set.seed(329) data <- data.frame( x1 = rnorm(500), x2 = rnorm(500), x3 = rnorm(500) ) data$y = with(data, expr = x1 + x2 + x3 + 1/2*x1 * x2 + x2 * x3 + rnorm(500)) forest <- orsf(data, y ~ ., n_tree = 5) orsf_vint(forest) #> interaction score #> 1: x2..x3 0.8021932 #> 2: x1..x2 0.5095065 #> 3: x1..x3 0.1133252"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":null,"dir":"Reference","previous_headings":"","what":"Variable selection — orsf_vs","title":"Variable selection — orsf_vs","text":"Variable selection","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Variable selection — orsf_vs","text":"","code":"orsf_vs(object, n_predictor_min = 3, verbose_progress = NULL)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Variable selection — orsf_vs","text":"object (ObliqueForest) trained oblique random forest object (see orsf). n_predictor_min (integer) minimum number predictors allowed verbose_progress (logical) implemented yet. progress printed console?","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Variable selection — orsf_vs","text":"data.table four columns: n_predictors: number predictors used stat_value: --bag statistic predictors_included: names predictors included predictor_dropped: predictor selected dropped","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Variable selection — orsf_vs","text":"tree_seeds specified object successive run orsf evaluated --bag samples initial run.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Variable selection — orsf_vs","text":"","code":"object <- orsf(formula = time + status ~ ., data = pbc_orsf, n_tree = 25, importance = 'anova') orsf_vs(object, n_predictor_min = 15) #> n_predictors stat_value predictors_included #> 1: 15 0.8356685 id,age,sex_f,ascites_1,spiders_1,edema_0.5,... #> 2: 16 0.8351997 id,age,sex_f,ascites_1,hepato_1,spiders_1,... #> 3: 17 0.8296786 id,trt_placebo,age,sex_f,ascites_1,hepato_1,... #> 4: 18 0.8185322 id,trt_placebo,age,sex_f,ascites_1,hepato_1,... #> predictor_dropped #> 1: platelet #> 2: hepato_1 #> 3: trt_placebo #> 4: alk.phos"},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":null,"dir":"Reference","previous_headings":"","what":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"data light modification survival::pbc data. modifications :","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"","code":"pbc_orsf"},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"data frame 276 rows 20 variables: id case number time number days registration earlier death, transplantion, study analysis July, 1986 status status endpoint, 0 censored transplant, 1 dead trt randomized treatment group: D-penicillmain placebo age years sex m/f ascites presence ascites hepato presence hepatomegaly enlarged liver spiders blood vessel malformations skin edema 0 edema, 0.5 untreated successfully treated, 1 edema despite diuretic therapy bili serum bilirubin (mg/dl) chol serum cholesterol (mg/dl) albumin serum albumin (g/dl) copper urine copper (ug/day) alk.phos alkaline phosphotase (U/liter) ast aspartate aminotransferase, called SGOT (U/ml) trig triglycerides (mg/dl) platelet platelet count protime standardized blood clotting time stage histologic stage disease (needs biopsy)","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"T Therneau P Grambsch (2000), Modeling Survival Data: Extending Cox Model, Springer-Verlag, New York. ISBN: 0-387-98784-3.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"removed rows missing data converted status 0 censor transplant, 1 dead converted stage ordered factor. converted trt, ascites, hepato, spiders, edema factors.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":null,"dir":"Reference","previous_headings":"","what":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"data copied lightly modified penguins data palmerpenguins R package. modification removal rows missing data. data include measurements penguin species, island Palmer Archipelago, size (flipper length, body mass, bill dimensions), sex.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"","code":"penguins_orsf"},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"tibble 333 rows 8 variables: species factor denoting penguin species (Adélie, Chinstrap Gentoo) island factor denoting island Palmer Archipelago, Antarctica (Biscoe, Dream Torgersen) bill_length_mm number denoting bill length (millimeters) bill_depth_mm number denoting bill depth (millimeters) flipper_length_mm integer denoting flipper length (millimeters) body_mass_g integer denoting body mass (grams) sex factor denoting penguin sex (female, male) year integer denoting study year (2007, 2008, 2009)","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"Adélie penguins: Palmer Station Antarctica LTER K. Gorman. 2020. Structural size measurements isotopic signatures foraging among adult male female Adélie penguins (Pygoscelis adeliae) nesting along Palmer Archipelago near Palmer Station, 2007-2009 ver 5. Environmental Data Initiative. doi:10.6073/pasta/98b16d7d563f265cb52372c8ca99e60f Gentoo penguins: Palmer Station Antarctica LTER K. Gorman. 2020. Structural size measurements isotopic signatures foraging among adult male female Gentoo penguin (Pygoscelis papua) nesting along Palmer Archipelago near Palmer Station, 2007-2009 ver 5. Environmental Data Initiative. doi:10.6073/pasta/7fca67fb28d56ee2ffa3d9370ebda689 Chinstrap penguins: Palmer Station Antarctica LTER K. Gorman. 2020. Structural size measurements isotopic signatures foraging among adult male female Chinstrap penguin (Pygoscelis antarcticus) nesting along Palmer Archipelago near Palmer Station, 2007-2009 ver 6. Environmental Data Initiative. doi:10.6073/pasta/c14dfcfada8ea13a17536e73eb6fbe9e Originally published : Gorman KB, Williams TD, Fraser WR (2014) Ecological Sexual Dimorphism Environmental Variability within Community Antarctic Penguins (Genus Pygoscelis). PLoS ONE 9(3): e90081. doi:10.1371/journal.pone.0090081","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":null,"dir":"Reference","previous_headings":"","what":"Automatic variable values for dependence — pred_spec_auto","title":"Automatic variable values for dependence — pred_spec_auto","text":"partial dependence individual conditional expectations, function allows variable considered without specify values set variable . values used based quantiles continuous variables (10th, 25th, 50th, 75th, 90th) unique categories categorical variables.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Automatic variable values for dependence — pred_spec_auto","text":"","code":"pred_spec_auto(...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Automatic variable values for dependence — pred_spec_auto","text":"... names variables use. can quotes quotes (see examples).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Automatic variable values for dependence — pred_spec_auto","text":"character vector names","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Automatic variable values for dependence — pred_spec_auto","text":"function used context orsf_pd orsf_ice functions.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Automatic variable values for dependence — pred_spec_auto","text":"","code":"fit <- orsf(penguins_orsf, species ~., n_tree = 5) orsf_pd_oob(fit, pred_spec_auto(flipper_length_mm)) #> class flipper_length_mm mean lwr medn upr #> 1: Adelie 185 0.5354109 0 0.50928468 1.0000000 #> 2: Adelie 190 0.5034360 0 0.47752809 1.0000000 #> 3: Adelie 197 0.4458775 0 0.11111111 1.0000000 #> 4: Adelie 213 0.3343601 0 0.07691303 1.0000000 #> 5: Adelie 221 0.3287078 0 0.07691303 1.0000000 #> 6: Chinstrap 185 0.2630916 0 0.03726345 1.0000000 #> 7: Chinstrap 190 0.2596916 0 0.03208729 1.0000000 #> 8: Chinstrap 197 0.2220773 0 0.02031115 1.0000000 #> 9: Chinstrap 213 0.2559835 0 0.03333333 1.0000000 #> 10: Chinstrap 221 0.2269255 0 0.02970297 0.9711538 #> 11: Gentoo 185 0.2014975 0 0.00000000 1.0000000 #> 12: Gentoo 190 0.2368724 0 0.00000000 1.0000000 #> 13: Gentoo 197 0.3320452 0 0.00000000 1.0000000 #> 14: Gentoo 213 0.4096564 0 0.14285714 1.0000000 #> 15: Gentoo 221 0.4443667 0 0.28571429 1.0000000"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":null,"dir":"Reference","previous_headings":"","what":"Prediction for ObliqueForest Objects — predict.ObliqueForest","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Compute predicted values oblique random forest. Predictions may returned aggregate (.e., averaging trees) tree-specific.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"","code":"# S3 method for ObliqueForest predict( object, new_data, pred_type = NULL, pred_horizon = NULL, pred_aggregate = TRUE, pred_simplify = FALSE, na_action = NULL, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"object (ObliqueForest) trained oblique random forest object (see orsf). new_data data.frame, tibble, data.table compute predictions . pred_type (character) type predictions compute. Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class 'class': predicted class regression: 'mean': predicted mean, .e., expected value pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_aggregate (logical) TRUE (default), predictions aggregated trees taking mean. FALSE, returned output contain one row per observation one column tree. length pred_horizon two pred_aggregate FALSE, result list matrices, 'th item list corresponding 'th value pred_horizon. pred_simplify (logical) FALSE (default), predictions always returned numeric matrix list numeric matrices. TRUE, predictions may simplified vector, e.g., pred_type 'mort' survival 'class' classification, array matrices length(pred_horizon) > 1. na_action (character) happen new_data contains missing values (.e., NA values). Valid options : 'fail' : error thrown new_data contains NA values 'pass' : output NA rows new_data 1 NA value predictors used object 'omit' : rows new_data incomplete data dropped 'impute_meanmode' : missing values continuous categorical variables new_data imputed using mean mode, respectively. clarify, mean mode used impute missing values training data object, new_data. boundary_checks (logical) TRUE, pred_horizon checked make sure requested values less maximum observed time object's training data. FALSE, checks skipped. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"matrix predictions. Column j matrix corresponds value j pred_horizon. Row matrix corresponds row new_data.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"new_data must columns equivalent types data used train object. Also, factors new_data must levels data used train object. pred_horizon values exceed maximum follow-time object's training data, truly want , set boundary_checks = FALSE can use pred_horizon large want. Note predictions beyond maximum follow-time object's training data equal predictions maximum follow-time, aorsf estimate survival beyond maximum observed time. unspecified, pred_horizon may automatically specified value used oobag_pred_horizon object created (see orsf).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"","code":"library(aorsf)"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"classification","dir":"Reference","previous_headings":"","what":"Classification","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Predict probability class predicted class:","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) # predicted probabilities, the default predict(fit_clsf, new_data = penguins_orsf_test[1:5, ], pred_type = 'prob') ## Adelie Chinstrap Gentoo ## [1,] 0.9405286 0.04125900 0.018212368 ## [2,] 0.9628964 0.03459853 0.002505059 ## [3,] 0.9029383 0.08527806 0.011783605 ## [4,] 0.9301983 0.05180907 0.017992625 ## [5,] 0.7968234 0.16538539 0.037791201 # predicted class (as a matrix by default) predict(fit_clsf, new_data = penguins_orsf_test[1:5, ], pred_type = 'class') ## [,1] ## [1,] 1 ## [2,] 1 ## [3,] 1 ## [4,] 1 ## [5,] 1 # predicted class (as a factor if you use simplify) predict(fit_clsf, new_data = penguins_orsf_test[1:5, ], pred_type = 'class', pred_simplify = TRUE) ## [1] Adelie Adelie Adelie Adelie Adelie ## Levels: Adelie Chinstrap Gentoo"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"regression","dir":"Reference","previous_headings":"","what":"Regression","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Predict mean value outcome:","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) predict(fit_regr, new_data = penguins_orsf_test[1:5, ], pred_type = 'mean') ## [,1] ## [1,] 37.74136 ## [2,] 37.42367 ## [3,] 37.04598 ## [4,] 39.89602 ## [5,] 39.14848"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"survival","dir":"Reference","previous_headings":"","what":"Survival","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Begin fitting oblique survival random forest: Predict risk, survival, cumulative hazard one several times: Predict mortality, defined number events forest’s population observations characteristics like current observation. type prediction require specify prediction horizon","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) # predicted risk, the default predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'risk', pred_horizon = c(500, 1000, 1500)) ## [,1] [,2] [,3] ## [1,] 0.013648562 0.058393393 0.11184029 ## [2,] 0.003811413 0.026857586 0.04774151 ## [3,] 0.030548361 0.100600301 0.14847107 ## [4,] 0.040381075 0.169596943 0.27018952 ## [5,] 0.001484698 0.006663576 0.01337655 # predicted survival, i.e., 1 - risk predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'surv', pred_horizon = c(500, 1000, 1500)) ## [,1] [,2] [,3] ## [1,] 0.9863514 0.9416066 0.8881597 ## [2,] 0.9961886 0.9731424 0.9522585 ## [3,] 0.9694516 0.8993997 0.8515289 ## [4,] 0.9596189 0.8304031 0.7298105 ## [5,] 0.9985153 0.9933364 0.9866235 # predicted cumulative hazard function # (expected number of events for person i at time j) predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'chf', pred_horizon = c(500, 1000, 1500)) ## [,1] [,2] [,3] ## [1,] 0.015395388 0.067815817 0.14942956 ## [2,] 0.004022524 0.028740305 0.05424314 ## [3,] 0.034832754 0.127687156 0.20899732 ## [4,] 0.059978334 0.233048809 0.42562310 ## [5,] 0.001651365 0.007173177 0.01393016 predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'mort') ## [,1] ## [1,] 23.405016 ## [2,] 15.362916 ## [3,] 26.180648 ## [4,] 36.515629 ## [5,] 5.856674"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":null,"dir":"Reference","previous_headings":"","what":"Inspect Forest Parameters — print.ObliqueForest","title":"Inspect Forest Parameters — print.ObliqueForest","text":"Printing ORSF model tells : Linear combinations: identified? N observations: Number rows training data N events: Number events training data N trees: Number trees forest N predictors total: Total number columns predictor matrix N predictors per node: Number variables used linear combinations Average leaves per tree: proxy depth trees Min observations leaf: See leaf_min_obs orsf Min events leaf: See leaf_min_events orsf OOB stat value: --bag error fitting trees OOB stat type: --bag error computed? Variable importance: variable importance computed?","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Inspect Forest Parameters — print.ObliqueForest","text":"","code":"# S3 method for ObliqueForest print(x, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Inspect Forest Parameters — print.ObliqueForest","text":"x (ObliqueForest) oblique random survival forest (ORSF; see orsf). ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Inspect Forest Parameters — print.ObliqueForest","text":"x, invisibly.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Inspect Forest Parameters — print.ObliqueForest","text":"","code":"object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 5) print(object) #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 20.8 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.76 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":null,"dir":"Reference","previous_headings":"","what":"Print ORSF summary — print.orsf_summary_uni","title":"Print ORSF summary — print.orsf_summary_uni","text":"Print ORSF summary","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Print ORSF summary — print.orsf_summary_uni","text":"","code":"# S3 method for orsf_summary_uni print(x, n_variables = NULL, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Print ORSF summary — print.orsf_summary_uni","text":"x object class 'orsf_summary' n_variables number variables print ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Print ORSF summary — print.orsf_summary_uni","text":"invisibly, x","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Print ORSF summary — print.orsf_summary_uni","text":"","code":"object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 25) smry <- orsf_summarize_uni(object, n_variables = 2) print(smry) #> #> -- ascites (VI Rank: 1) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3087374 0.1859818 0.04183841 0.5614236 #> 1 0.4965395 0.4214549 0.30005793 0.7145065 #> #> -- bili (VI Rank: 2) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2404645 0.1342975 0.03456875 0.3904944 #> 0.80 0.2425520 0.1435297 0.03507037 0.3904944 #> 1.40 0.2625349 0.1554767 0.04820122 0.4168304 #> 3.52 0.3795922 0.3167143 0.15794919 0.5811623 #> 7.25 0.4682454 0.4348054 0.25161269 0.6726243 #> #> Predicted risk at time t = 1788 for top 2 predictors"},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-012-unreleased","dir":"Changelog","previous_headings":"","what":"aorsf 0.1.2 (unreleased)","title":"aorsf 0.1.2 (unreleased)","text":"CRAN release: 2024-01-15 Allowed option \"time\" pred_type predict partial dependence predict survival time (see https://github.com/ropensci/aorsf/issues/37). Added pred_spec_auto() convenient specification variables partial dependence. Partial dependence now runs much faster multiple threads. Added orsf_vint() compute variable interaction scores using partial dependence. Added orsf_update(), can copy modify obliqueForest modify place. Added orsf_control functions classification, regression, survival (https://github.com/ropensci/aorsf/pull/25). optimization implemented matrix multiplication prediction (https://github.com/ropensci/aorsf/pull/20)","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-011","dir":"Changelog","previous_headings":"","what":"aorsf 0.1.1","title":"aorsf 0.1.1","text":"CRAN release: 2023-10-26 Fixed uninitialized value pd_type Fixed various issues related memory leaks","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-010","dir":"Changelog","previous_headings":"","what":"aorsf 0.1.0","title":"aorsf 0.1.0","text":"CRAN release: 2023-10-13 Re-worked internal C++ routines following design ranger. Re-worked progress printed console verbose_progress TRUE, following design ranger. Messages now indicate action taken, % complete, approximate time finishing action. Improved variable importance, following design ranger. Importance now computed tree--tree instead aggregate. Additionally, mortality type prediction used importance survival trees, since mortality depend pred_horizon. Allowed multi-threading performed orsf(), predict.orsf_fit(), functions orsf_vi() orsf_pd() family. Allowed sampling without replacement sampling specific fraction observations orsf() Included Harrell’s C-statistic option assessing goodness splits growing trees. Fixed issue uninformative error message occur pred_horizon > max(time) orsf_summarize_uni. Thanks @JyHao1 @DustinMLong finding !","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-007","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.7","title":"aorsf 0.0.7","text":"CRAN release: 2023-01-12 Additional changes internal testing avoid problems ATLAS","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-006","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.6","title":"aorsf 0.0.6","text":"CRAN release: 2023-01-06 Minor fix internal tests failing run ATLAS","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-005","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.5","title":"aorsf 0.0.5","text":"CRAN release: 2022-12-14 orsf() longer throws errors warnings try give single predictor. note added documentation details ?orsf explains using single predictor orsf() somewhat useless. done resolve https://github.com/mlr-org/mlr3extralearners/issues/259. predict.orsf_fit now accepts pred_horizon = 0 returns sensible values. Thanks @mattwarkentin feature request. added function perform variable selection, orsf_vs(). Made variable importance consistent respect group_factors. Originally, output orsf ungrouped VI values orsf_vi grouped values. update, orsf defaults grouped values. ungrouped values can still recovered. Fixed issue orsf_pd functions output data returned original scale.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-004","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.4","title":"aorsf 0.0.4","text":"CRAN release: 2022-11-07 orsf formulas now accepts Surv objects (see https://github.com/ropensci/aorsf/issues/11) Added verbose_progress input orsf, prints messages console indicating progress. Allowance missing values orsf. Mean mode imputation performed observations missing data. values can also used impute new data missing values. Centering scaling predictors now done prior growing forest.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-003","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.3","title":"aorsf 0.0.3","text":"CRAN release: 2022-10-09 Included rOpenSci reviewers Christopher Jackson, Marvin N Wright, Lukas Burk DESCRIPTION reviewers. Thank ! Added clarification docs pros/cons different variable importance techniques Added regression tests aorsf versus obliqueRSF (similar) Additional support tests functions long right hand sides Updated --bag vignette appropriate custom functions. Allow status values input data general, .e., just 0 1. Allow missing values predict functions, including partial dependence.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-002","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.2","title":"aorsf 0.0.2","text":"CRAN release: 2022-09-05 Modified unit tests compatibility extra checks run CRAN.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-001","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.1","title":"aorsf 0.0.1","text":"CRAN release: 2022-08-23 Added orsf_control_custom(), allows users submit custom functions identifying linear combinations inputs growing oblique decision trees. Added weights input orsf, allowing users fit orsf specific data training set. Added chf mort options predict.orsf_fit(). Mortality predictions fully implemented yet - supported partial dependence --bag error estimates. features added future update.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-0009000","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.0.9000","title":"aorsf 0.0.0.9000","text":"Core features implemented: fit, interpret, predict using oblique random survival forests. Vignettes + Readme covering usage core features. Website hosted GitHub pages, managed pkgdown.","code":""}] +[{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":null,"dir":"","previous_headings":"","what":"Contributing to aorsf","title":"Contributing to aorsf","text":"Want contribute aorsf? Great! aorsf initially stable state development, great deal active subsequent development envisioned. outline propose change aorsf. detailed info contributing , tidyverse packages, please see development contributing guide.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"fixing-typos","dir":"","previous_headings":"","what":"Fixing typos","title":"Contributing to aorsf","text":"can fix typos, spelling mistakes, grammatical errors documentation directly using GitHub web interface, long changes made source file. generally means ’ll need edit roxygen2 comments .R, .Rd file. can find .R file generates .Rd reading comment first line.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"bigger-changes","dir":"","previous_headings":"","what":"Bigger changes","title":"Contributing to aorsf","text":"want make bigger change, ’s good idea first file issue make sure someone team agrees ’s needed. ’ve found bug, please file issue illustrates bug minimal reprex (also help write unit test, needed).","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"pull-request-process","dir":"","previous_headings":"Bigger changes","what":"Pull request process","title":"Contributing to aorsf","text":"Fork package clone onto computer. haven’t done , recommend using usethis::create_from_github(\"ropensci/aorsf\", fork = TRUE). Install development dependencies devtools::install_dev_deps(), make sure package passes R CMD check running devtools::check(). R CMD check doesn’t pass cleanly, ’s good idea ask help continuing. Create Git branch pull request (PR). recommend using usethis::pr_init(\"brief-description--change\"). Make changes, commit git, create PR running usethis::pr_push(), following prompts browser. title PR briefly describe change. body PR contain Fixes #issue-number. user-facing changes, add bullet top NEWS.md (.e. just first header). Follow style described https://style.tidyverse.org/news.html.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"code-style","dir":"","previous_headings":"Bigger changes","what":"Code style","title":"Contributing to aorsf","text":"New code follow tidyverse style guide. can use styler package apply styles, please don’t restyle code nothing PR. use roxygen2, Markdown syntax, documentation. use testthat unit tests. Contributions test cases included easier accept.","code":""},{"path":"https://bcjaeger.github.io/aorsf/CONTRIBUTING.html","id":"code-of-conduct","dir":"","previous_headings":"","what":"Code of Conduct","title":"Contributing to aorsf","text":"Please note aorsf project released Contributor Code Conduct. contributing project agree abide terms.","code":""},{"path":"https://bcjaeger.github.io/aorsf/LICENSE.html","id":null,"dir":"","previous_headings":"","what":"MIT License","title":"MIT License","text":"Copyright (c) 2022 aorsf authors (Byron C. Jaeger, Sawyer Welden, Nicholas M. Pajewski) Permission hereby granted, free charge, person obtaining copy software associated documentation files (“Software”), deal Software without restriction, including without limitation rights use, copy, modify, merge, publish, distribute, sublicense, /sell copies Software, permit persons Software furnished , subject following conditions: copyright notice permission notice shall included copies substantial portions Software. SOFTWARE PROVIDED “”, WITHOUT WARRANTY KIND, EXPRESS IMPLIED, INCLUDING LIMITED WARRANTIES MERCHANTABILITY, FITNESS PARTICULAR PURPOSE NONINFRINGEMENT. EVENT SHALL AUTHORS COPYRIGHT HOLDERS LIABLE CLAIM, DAMAGES LIABILITY, WHETHER ACTION CONTRACT, TORT OTHERWISE, ARISING , CONNECTION SOFTWARE USE DEALINGS SOFTWARE.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"background","dir":"Articles","previous_headings":"","what":"Background","title":"Introduction to aorsf","text":"oblique random forest (RF) extension traditional (axis-based) RF. Instead using single variable split data grow new branches, trees oblique RF use weighted combination multiple variables.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"oblique-rfs-for-survival-classification-and-regression","dir":"Articles","previous_headings":"","what":"Oblique RFs for survival, classification, and regression","title":"Introduction to aorsf","text":"purpose aorsf (‘’ short accelerated) provide unifying framework fit oblique RFs can scale adequately large data sets. fastest algorithms available package used default often equivalent prediction accuracy computational approaches. center piece aorsf orsf() function. initial versions aorsf, orsf() function fit oblique random survival forests, now allows classification, regression, survival forests. (may introduce orf() function future name orsf() misleading users.) classification, fit oblique RF predict penguin species using penguin data magnificent palmerpenguins R package regression, use data predict bill length penguins: personal favorite oblique survival RF accelerated Cox regression great combination prediction accuracy computational efficiency (see JCGS paper). , predict mortality risk following diagnosis primary biliary cirrhosis: may notice first input aorsf data. design choice makes easier use orsf pipes (.e., %>% |>). instance,","code":"# An oblique classification RF penguin_fit <- orsf(data = penguins_orsf, formula = species ~ .) penguin_fit #> ---------- Oblique random classification forest #> #> Linear combinations: Accelerated Logistic regression #> N observations: 333 #> N classes: 3 #> N trees: 500 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 5.544 #> Min observations in leaf: 5 #> OOB stat value: 1.00 #> OOB stat type: AUC-ROC #> Variable importance: anova #> #> ----------------------------------------- # An oblique regression RF bill_fit <- orsf(data = penguins_orsf, formula = bill_length_mm ~ .) bill_fit #> ---------- Oblique random regression forest #> #> Linear combinations: Accelerated Linear regression #> N observations: 333 #> N trees: 500 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 49.872 #> Min observations in leaf: 5 #> OOB stat value: 0.81 #> OOB stat type: RSQ #> Variable importance: anova #> #> ----------------------------------------- # An oblique survival RF pbc_fit <- orsf(data = pbc_orsf, n_tree = 5, formula = Surv(time, status) ~ . - id) pbc_fit #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 21.4 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.77 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> ----------------------------------------- library(dplyr) pbc_fit <- pbc_orsf |> select(-id) |> orsf(formula = Surv(time, status) ~ ., n_tree = 5)"},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"interpretation","dir":"Articles","previous_headings":"","what":"Interpretation","title":"Introduction to aorsf","text":"aorsf includes several functions dedicated interpretation ORSFs, estimation partial dependence variable importance.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"variable-importance","dir":"Articles","previous_headings":"Interpretation","what":"Variable importance","title":"Introduction to aorsf","text":"multiple methods compute variable importance, can applied type oblique forest. compute negation importance, ORSF multiplies coefficient variable -1 re-computes --sample (sometimes referred --bag) accuracy ORSF model. can also compute variable importance using permutation, classical approach noises predictor assigned resulting degradation prediction accuracy importance predictor. faster alternative permutation negation importance ANOVA importance, computes proportion times variable obtains low p-value (p < 0.01) forest grown.","code":"orsf_vi_negate(pbc_fit) #> bili spiders copper albumin sex #> 0.1671647745 0.0520379071 0.0432409174 0.0292245433 0.0258174901 #> alk.phos trt protime ast stage #> 0.0207262097 0.0196654076 0.0147894928 0.0138953587 0.0094377622 #> age ascites trig edema platelet #> 0.0063243985 0.0053495205 0.0021405459 0.0018460367 0.0008203822 #> hepato chol #> 0.0004910894 -0.0128635357 orsf_vi_permute(penguin_fit) #> bill_length_mm flipper_length_mm bill_depth_mm island #> 0.1738256958 0.0898851835 0.0810334484 0.0655328821 #> body_mass_g sex year #> 0.0647936728 0.0191948220 0.0005500798 orsf_vi_anova(bill_fit) #> species sex island flipper_length_mm #> 0.35859861 0.20116912 0.11896429 0.08734940 #> body_mass_g bill_depth_mm year #> 0.08107428 0.06459390 0.01368270"},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"partial-dependence-pd","dir":"Articles","previous_headings":"Interpretation","what":"Partial dependence (PD)","title":"Introduction to aorsf","text":"Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model’s prediction. PD, see vignette","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"individual-conditional-expectations-ice","dir":"Articles","previous_headings":"Interpretation","what":"Individual conditional expectations (ICE)","title":"Introduction to aorsf","text":"Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor. ICE, see vignette","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"what-about-the-original-orsf","dir":"Articles","previous_headings":"","what":"What about the original ORSF?","title":"Introduction to aorsf","text":"original ORSF (.e., obliqueRSF) used glmnet find linear combinations inputs. aorsf allows users implement approach using orsf_control_survival(method = 'net') function: net forests fit lot faster original ORSF function obliqueRSF. However, net forests still much slower cph ones.","code":"orsf_net <- orsf(data = pbc_orsf, formula = Surv(time, status) ~ . - id, control = orsf_control_survival(method = 'net'))"},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"aorsf-and-other-machine-learning-software","dir":"Articles","previous_headings":"","what":"aorsf and other machine learning software","title":"Introduction to aorsf","text":"unique feature aorsf fast algorithms fit ORSF ensembles. RLT obliqueRSF fit oblique random survival forests, aorsf faster. ranger randomForestSRC fit survival forests, neither package supports oblique splitting. obliqueRF fits oblique random forests classification regression, survival. PPforest fits oblique random forests classification survival. Note: default prediction behavior aorsf models produce predicted risk specific prediction horizon, default ranger randomForestSRC. think change future, computing time independent predictions aorsf helpful.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/aorsf.html","id":"learning-more","dir":"Articles","previous_headings":"","what":"Learning more","title":"Introduction to aorsf","text":"aorsf began dedicated package oblique random survival forests, papers published far focused survival analysis risk prediction. However, routines regression classification oblique RFs aorsf high overlap survival ones. See orsf details oblique random survival forests. see JCGS paper details algorithms used specifically aorsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"go-faster","dir":"Articles","previous_headings":"","what":"Go faster","title":"Tips to speed up computation","text":"Analyses can slow crawl models need hours run. article find tricks prevent bottleneck using orsf().","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"dont-specify-a-control","dir":"Articles","previous_headings":"","what":"Don’t specify a control","title":"Tips to speed up computation","text":"default control orsf() NULL , unspecified, orsf() pick fastest possible control depending type forest grown. default control run-time compared approaches can striking. example:","code":"time_fast <- system.time( expr = orsf(pbc_orsf, formula = time+status~. -id, n_tree = 5) ) time_net <- system.time( expr = orsf(pbc_orsf, formula = time+status~. -id, control = orsf_control_survival(method = 'net'), n_tree = 5) ) # unspecified control is much faster time_net['elapsed'] / time_fast['elapsed'] #> elapsed #> 44.125"},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"use-n_thread","dir":"Articles","previous_headings":"","what":"Use n_thread","title":"Tips to speed up computation","text":"n_thread argument uses multi-threading run aorsf functions parallel possible. know many threads want, e.g. want exactly 5, set n_thread = 5. aren’t sure many threads available want use feasible amount, using n_thread = 0 (default) tells aorsf . Note: sometimes multi-threading possible. example, R single threaded language, multi-threading applied orsf() needs call R functions C++, occurs customized R function used find linear combination variables compute prediction accuracy.","code":"# automatically pick number of threads based on amount available orsf(pbc_orsf, formula = time+status~. -id, n_tree = 5, n_thread = 0)"},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"do-less","dir":"Articles","previous_headings":"","what":"Do less","title":"Tips to speed up computation","text":"inputs orsf() can adjusted make run faster: set n_retry 0 set oobag_pred_type 'none' set importance 'none' increase split_min_events, split_min_obs, leaf_min_events, leaf_min_obs make trees stop growing sooner increase split_min_stat enforce strict requirements growing deeper trees. Applying tips: modifying inputs can make orsf() run faster, can also impact prediction accuracy.","code":"orsf(pbc_orsf, formula = time+status~., n_thread = 0, n_tree = 5, n_retry = 0, oobag_pred_type = 'none', importance = 'none', split_min_events = 20, leaf_min_events = 10, split_min_stat = 10)"},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"show-progress","dir":"Articles","previous_headings":"","what":"Show progress","title":"Tips to speed up computation","text":"Setting verbose_progress = TRUE doesn’t make anything run faster, can help make feel like things running less slow.","code":"verbose_fit <- orsf(pbc_orsf, formula = time+status~. -id, n_tree = 5, verbose_progress = TRUE) #> Growing trees: 100%. #> Computing predictions: 100%."},{"path":"https://bcjaeger.github.io/aorsf/articles/fast.html","id":"dont-wait--estimate","dir":"Articles","previous_headings":"","what":"Don’t wait. Estimate!","title":"Tips to speed up computation","text":"Instead running model hoping fast, can estimate long specification model take using no_fit = TRUE call orsf().","code":"fit_spec <- orsf(pbc_orsf, formula = time+status~. -id, control = orsf_control_survival(method = 'net'), n_tree = 2000, no_fit = TRUE) # how much time it takes to estimate training time: system.time( time_est <- orsf_time_to_train(fit_spec, n_tree_subset = 5) ) #> user system elapsed #> 0.286 0.020 0.307 # the estimated training time: time_est #> Time difference of 122.3736 secs"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"out-of-bag-data","dir":"Articles","previous_headings":"","what":"Out-of-bag data","title":"Out-of-bag predictions and evaluation","text":"random forests, tree grown bootstrapped version training set. bootstrap samples selected replacement, bootstrapped training set contains two-thirds instances original training set. ‘--bag’ data instances bootstrapped training set.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"out-of-bag-predictions-and-error","dir":"Articles","previous_headings":"","what":"Out-of-bag predictions and error","title":"Out-of-bag predictions and evaluation","text":"tree random forest can make predictions --bag data, --bag predictions can aggregated make ensemble --bag prediction. Since --bag data used grow tree, accuracy ensemble --bag predictions approximate generalization error random forest. --bag prediction error plays central role routines estimate variable importance, e.g. negation importance. fit oblique random survival forest plot distribution ensemble --bag predictions. Next, let’s check --bag accuracy fit: --bag estimate Harrell’s C-index (default method evaluate --bag predictions) 0.7652389.","code":"fit <- orsf(data = pbc_orsf, formula = Surv(time, status) ~ . - id, oobag_pred_type = 'surv', n_tree = 5, oobag_pred_horizon = 2000) hist(fit$pred_oobag, main = 'Out-of-bag survival predictions at t=2,000') # what function is used to evaluate out-of-bag predictions? fit$eval_oobag$stat_type #> [1] \"Harrell's C-index\" # what is the output from this function? fit$eval_oobag$stat_values #> [,1] #> [1,] 0.7652389"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"monitoring-out-of-bag-error","dir":"Articles","previous_headings":"","what":"Monitoring out-of-bag error","title":"Out-of-bag predictions and evaluation","text":"--bag data set contains one-third training set, --bag error estimate usually converges stable value trees added forest. want monitor convergence --bag error oblique random survival forest, can set oobag_eval_every compute --bag error every oobag_eval_every tree. example, let’s compute --bag error fitting tree forest 50 trees: general, least 500 trees recommended random forest fit. ’re just using 10 illustration.","code":"fit <- orsf(data = pbc_orsf, formula = Surv(time, status) ~ . - id, n_tree = 20, tree_seeds = 2, oobag_pred_type = 'surv', oobag_pred_horizon = 2000, oobag_eval_every = 1) plot( x = seq(1, 20, by = 1), y = fit$eval_oobag$stat_values, main = 'Out-of-bag C-statistic computed after each new tree is grown.', xlab = 'Number of trees grown', ylab = fit$eval_oobag$stat_type ) lines(x=seq(1, 20), y = fit$eval_oobag$stat_values)"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"user-supplied-out-of-bag-evaluation-functions","dir":"Articles","previous_headings":"","what":"User-supplied out-of-bag evaluation functions","title":"Out-of-bag predictions and evaluation","text":"cases, may want control --bag error estimated. example, let’s use Brier score SurvMetrics package: two ways apply function compute --bag error. First, can apply function --bag survival predictions stored ‘aorsf’ objects, e.g: Second, can pass function orsf(), used place Harrell’s C-statistic:","code":"oobag_brier_surv <- function(y_mat, w_vec, s_vec){ # output is numeric vector of length 1 as.numeric( SurvMetrics::Brier( object = Surv(time = y_mat[, 1], event = y_mat[, 2]), pre_sp = s_vec, # t_star in Brier() should match oob_pred_horizon in orsf() t_star = 2000 ) ) } oobag_brier_surv(y_mat = pbc_orsf[,c('time', 'status')], s_vec = fit$pred_oobag) #> [1] 0.11869 # instead of copy/pasting the modeling code and then modifying it, # you can just use orsf_update. fit_brier <- orsf_update(fit, oobag_fun = oobag_brier_surv) plot( x = seq(1, 20, by = 1), y = fit_brier$eval_oobag$stat_values, main = 'Out-of-bag error computed after each new tree is grown.', sub = 'For the Brier score, lower values indicate more accurate predictions', xlab = 'Number of trees grown', ylab = \"Brier score\" ) lines(x=seq(1, 20), y = fit_brier$eval_oobag$stat_values)"},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"specific-instructions-on-user-supplied-functions","dir":"Articles","previous_headings":"User-supplied out-of-bag evaluation functions","what":"Specific instructions on user-supplied functions","title":"Out-of-bag predictions and evaluation","text":"use oobag_fun note following: oobag_fun three inputs: y_mat, w_vec, s_vec survival trees, y_mat two column matrix first column named ‘time’ second named ‘status’. classification trees, y_mat matrix number columns = number distinct classes outcome. regression, y_mat matrix one column. s_vec numeric vector containing predictions oobag_fun return numeric output length 1","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/oobag.html","id":"notes","dir":"Articles","previous_headings":"","what":"Notes","title":"Out-of-bag predictions and evaluation","text":"evaluating --bag error: oobag_pred_horizon input orsf() determines prediction horizon --bag predictions. prediction horizon needs specified evaluate prediction accuracy cases, examples . sure check case using functions, , sure oobag_pred_horizon matches prediction horizon used custom function. functions expect predicted risk (.e., 1 - predicted survival), others expect predicted survival.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"partial-dependence-pd","dir":"Articles","previous_headings":"","what":"Partial dependence (PD)","title":"PD and ICE curves with ORSF","text":"Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model’s prediction. can compute PD individual conditional expectation (ICE) three ways: using -bag predictions training data. -bag PD indicates relationships model learned training. helpful goal interpret model. using --bag predictions training data. --bag PD indicates relationships model learned training using --bag data simulates application model new data. helpful want test model’s reliability fairness new data don’t access large testing set. using predictions new set data. New data PD shows model predicts outcomes observations seen. helpful want test model’s reliability fairness.","code":"library(aorsf) library(ggplot2)"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"classification","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Classification","title":"PD and ICE curves with ORSF","text":"Begin fitting oblique classification random forest: Compute PD using --bag data flipper_length_mm = c(190, 210). Note predicted probabilities returned class probabilities mean column sum 1 take sum class specific value pred_spec variables. example, isn’t case median predicted probability!","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_oob <- orsf_pd_oob(fit_clsf, pred_spec = pred_spec) pd_oob #> class flipper_length_mm mean lwr medn upr #> 1: Adelie 190 0.6182417 0.206899034 0.75537171 0.9796439 #> 2: Adelie 210 0.4348386 0.019519733 0.56802082 0.8620694 #> 3: Chinstrap 190 0.2114905 0.018420139 0.15561560 0.7174734 #> 4: Chinstrap 210 0.1806274 0.020409141 0.09928047 0.6990198 #> 5: Gentoo 190 0.1702678 0.001281382 0.02830728 0.5733438 #> 6: Gentoo 210 0.3845340 0.072260715 0.20258335 0.9519486 sum(pd_oob[flipper_length_mm == 190, mean]) #> [1] 1 sum(pd_oob[flipper_length_mm == 190, medn]) #> [1] 0.9392946"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"regression","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Regression","title":"PD and ICE curves with ORSF","text":"Begin fitting oblique regression random forest: Compute PD using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new #> flipper_length_mm mean lwr medn upr #> 1: 190 42.96571 37.09805 43.69769 48.72301 #> 2: 210 45.66012 40.50693 46.31577 51.65163 pred_spec = pred_spec_auto(species, island, body_mass_g) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new #> species island body_mass_g mean lwr medn upr #> 1: Adelie Biscoe 3200 40.31374 37.24373 40.31967 44.22824 #> 2: Chinstrap Biscoe 3200 45.10582 42.63342 45.10859 47.60119 #> 3: Gentoo Biscoe 3200 42.81649 40.19221 42.55664 46.84035 #> 4: Adelie Dream 3200 40.16219 36.95895 40.34633 43.90681 #> 5: Chinstrap Dream 3200 46.21778 43.53954 45.90929 49.19173 #> 6: Gentoo Dream 3200 42.60465 39.89647 42.63520 46.28769 #> 7: Adelie Torgersen 3200 39.91652 36.80227 39.79806 43.68842 #> 8: Chinstrap Torgersen 3200 44.27807 41.95470 44.40742 46.68848 #> 9: Gentoo Torgersen 3200 42.09510 39.49863 41.80049 45.81833 #> 10: Adelie Biscoe 3550 40.77971 38.04027 40.59561 44.57505 #> 11: Chinstrap Biscoe 3550 45.81304 43.52102 45.73116 48.36366 #> 12: Gentoo Biscoe 3550 43.31233 40.77355 43.03077 47.22936 #> 13: Adelie Dream 3550 40.77741 38.07399 40.78175 44.37273 #> 14: Chinstrap Dream 3550 47.30926 44.80493 46.77540 50.47092 #> 15: Gentoo Dream 3550 43.26955 40.86119 43.16204 46.89190 #> 16: Adelie Torgersen 3550 40.25780 37.35251 40.07871 44.04576 #> 17: Chinstrap Torgersen 3550 44.77911 42.60161 44.81944 47.14986 #> 18: Gentoo Torgersen 3550 42.49520 39.95866 42.14160 46.26237 #> 19: Adelie Biscoe 3975 41.61744 38.94515 41.36634 45.38752 #> 20: Chinstrap Biscoe 3975 46.59363 44.59970 46.44923 49.11457 #> 21: Gentoo Biscoe 3975 44.07857 41.60792 43.74562 47.85109 #> 22: Adelie Dream 3975 41.50511 39.06187 41.24741 45.13027 #> 23: Chinstrap Dream 3975 48.14978 45.87390 47.54867 51.50683 #> 24: Gentoo Dream 3975 44.01928 41.70577 43.84099 47.50470 #> 25: Adelie Torgersen 3975 40.94764 38.12519 40.66759 44.73689 #> 26: Chinstrap Torgersen 3975 45.44820 43.49986 45.44036 47.63243 #> 27: Gentoo Torgersen 3975 43.13791 40.70628 42.70627 46.87306 #> 28: Adelie Biscoe 4700 42.93914 40.48463 42.44768 46.81756 #> 29: Chinstrap Biscoe 4700 47.18517 45.40866 47.07739 49.55747 #> 30: Gentoo Biscoe 4700 45.32541 43.08173 44.93498 49.23391 #> 31: Adelie Dream 4700 42.73806 40.44229 42.22226 46.49936 #> 32: Chinstrap Dream 4700 48.37278 46.34335 48.00781 51.18955 #> 33: Gentoo Dream 4700 45.09132 42.88328 44.79530 48.82180 #> 34: Adelie Torgersen 4700 42.09349 39.72074 41.56168 45.68838 #> 35: Chinstrap Torgersen 4700 46.16807 44.38410 46.09525 48.35127 #> 36: Gentoo Torgersen 4700 44.31621 42.18968 43.81773 47.98024 #> 37: Adelie Biscoe 5300 43.89769 41.43335 43.28504 48.10892 #> 38: Chinstrap Biscoe 5300 47.53721 45.66038 47.52770 49.88701 #> 39: Gentoo Biscoe 5300 46.16115 43.81722 45.59309 50.57469 #> 40: Adelie Dream 5300 43.59846 41.25825 43.24518 47.46193 #> 41: Chinstrap Dream 5300 48.48139 46.36282 48.25679 51.02996 #> 42: Gentoo Dream 5300 45.91819 43.62832 45.54110 49.91622 #> 43: Adelie Torgersen 5300 42.92879 40.66576 42.31072 46.76406 #> 44: Chinstrap Torgersen 5300 46.59576 44.80400 46.49196 49.03906 #> 45: Gentoo Torgersen 5300 45.11384 42.95190 44.51289 49.27629 #> species island body_mass_g mean lwr medn upr pd_new <- orsf_pd_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new #> variable value level mean lwr medn upr #> 1: species NA Adelie 41.90271 37.10417 41.51723 48.51478 #> 2: species NA Chinstrap 47.11314 42.40419 46.96478 51.51392 #> 3: species NA Gentoo 44.37038 39.87306 43.89889 51.21635 #> 4: island NA Biscoe 44.21332 37.22711 45.27862 51.21635 #> 5: island NA Dream 44.43354 37.01471 45.57261 51.51392 #> 6: island NA Torgersen 43.29539 37.01513 44.26924 49.84391 #> 7: body_mass_g 3200 42.84625 37.03978 43.95991 49.19173 #> 8: body_mass_g 3550 43.53326 37.56730 44.43756 50.47092 #> 9: body_mass_g 3975 44.30431 38.31567 45.22089 51.50683 #> 10: body_mass_g 4700 45.22525 39.88199 46.34680 51.18955 #> 11: body_mass_g 5300 45.91412 40.84742 46.95327 51.48851 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') pd_new <- orsf_pd_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) pd_new #> species island mean lwr medn upr #> 1: Adelie Biscoe 41.98024 37.22711 41.65252 48.51478"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"survival","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Survival","title":"PD and ICE curves with ORSF","text":"Begin fitting oblique survival random forest: Compute PD using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get PD value:","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) pd_train <- orsf_pd_inb(fit_surv, pred_spec = list(bili = 1:5)) pd_train #> pred_horizon bili mean lwr medn upr #> 1: 1826.25 1 0.2575450 0.02234786 0.1334170 0.8917942 #> 2: 1826.25 2 0.3130469 0.06853733 0.1906695 0.9203372 #> 3: 1826.25 3 0.3711963 0.11409793 0.2582027 0.9416791 #> 4: 1826.25 4 0.4248968 0.15648381 0.3334579 0.9591581 #> 5: 1826.25 5 0.4671699 0.20123406 0.3855137 0.9655296 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili)) pd_train #> pred_horizon bili mean lwr medn upr #> 1: 1826.25 0.590 0.2493753 0.02035041 0.1250263 0.8823385 #> 2: 1826.25 0.725 0.2517103 0.02060111 0.1281814 0.8836536 #> 3: 1826.25 1.500 0.2807082 0.03964900 0.1601715 0.9040617 #> 4: 1826.25 3.500 0.3968251 0.13431288 0.2934565 0.9501230 #> 5: 1826.25 7.210 0.5352155 0.27869513 0.4658256 0.9782084 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) pd_train #> pred_horizon bili mean lwr medn upr #> 1: 500 0.590 0.06217164 0.0004433990 0.008765301 0.5918852 #> 2: 1000 0.590 0.14282695 0.0057937418 0.056509484 0.7381953 #> 3: 1500 0.590 0.20944972 0.0136094784 0.092379507 0.8577223 #> 4: 2000 0.590 0.26917477 0.0230476894 0.146421502 0.8918696 #> 5: 2500 0.590 0.31901518 0.0631155452 0.203673185 0.9034059 #> 6: 3000 0.590 0.39244000 0.0911566314 0.302726475 0.9239494 #> 7: 500 0.725 0.06287876 0.0004462367 0.009001904 0.5980510 #> 8: 1000 0.725 0.14409310 0.0063321712 0.056833294 0.7448126 #> 9: 1500 0.725 0.21143724 0.0140736894 0.093685200 0.8597396 #> 10: 2000 0.725 0.27150368 0.0235448705 0.147022224 0.8940497 #> 11: 2500 0.725 0.32014805 0.0626303822 0.203946002 0.9073003 #> 12: 3000 0.725 0.39518173 0.0911457406 0.308428469 0.9252028 #> 13: 500 1.500 0.06712295 0.0012717884 0.011028398 0.6240769 #> 14: 1000 1.500 0.15802582 0.0114789623 0.068332010 0.7683888 #> 15: 1500 1.500 0.23407183 0.0287320952 0.117289745 0.8789647 #> 16: 2000 1.500 0.30235436 0.0467927208 0.180096425 0.9143235 #> 17: 2500 1.500 0.35354874 0.0845866747 0.238415966 0.9265099 #> 18: 3000 1.500 0.43604287 0.1311103304 0.348078730 0.9438196 #> 19: 500 3.500 0.08677320 0.0052087533 0.028244374 0.6741102 #> 20: 1000 3.500 0.22427808 0.0519179775 0.139857107 0.8277541 #> 21: 1500 3.500 0.32788654 0.0901983241 0.217982772 0.9371150 #> 22: 2000 3.500 0.41708208 0.1445328597 0.313224605 0.9566091 #> 23: 2500 3.500 0.49334883 0.2195110942 0.402932569 0.9636221 #> 24: 3000 3.500 0.56094391 0.2647541788 0.503509668 0.9734948 #> 25: 500 7.210 0.12591911 0.0220920570 0.063283130 0.7522611 #> 26: 1000 7.210 0.32642477 0.1353851175 0.259731888 0.8879218 #> 27: 1500 7.210 0.46409472 0.2181840827 0.387142510 0.9700903 #> 28: 2000 7.210 0.55116942 0.2912654769 0.484118150 0.9811496 #> 29: 2500 7.210 0.62008114 0.3709845684 0.568822502 0.9844945 #> 30: 3000 7.210 0.68030697 0.4247511750 0.646009789 0.9888637 #> pred_horizon bili mean lwr medn upr"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"one-variable-moving-horizon","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"One variable, moving horizon","title":"PD and ICE curves with ORSF","text":"next sections, update orsf_fit include data pbc_orsf instead just training sample: effect predictor varies time? Partial dependence can show . inspection, can see males higher risk females difference risk grows time. can also seen viewing ratio expected risk time: get view PD number variables training data, use orsf_summarize_uni(). function computes --bag PD important n_variables returns nicely formatted view output: ‘summary’ object can converted data.table downstream plotting tables.","code":"# a rare case of modify_in_place = TRUE orsf_update(fit_surv, data = pbc_orsf, modify_in_place = TRUE) fit_surv #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 500 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 21.038 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.84 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> ----------------------------------------- pd_sex_tv <- orsf_pd_oob(fit_surv, pred_spec = pred_spec_auto(sex), pred_horizon = seq(365, 365*5)) ggplot(pd_sex_tv) + aes(x = pred_horizon, y = mean, color = sex) + geom_line() + labs(x = 'Time since baseline', y = 'Expected risk') library(data.table) ratio_tv <- pd_sex_tv[ , .(ratio = mean[sex == 'm'] / mean[sex == 'f']), by = pred_horizon ] ggplot(ratio_tv, aes(x = pred_horizon, y = ratio)) + geom_line(color = 'grey') + geom_smooth(color = 'black', se = FALSE) + labs(x = 'time since baseline', y = 'ratio in expected risk for males versus females') pd_smry <- orsf_summarize_uni(fit_surv, n_variables = 4) pd_smry #> #> -- ascites (VI Rank: 1) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3083328 0.1985589 0.06581247 0.5241336 #> 1 0.4702396 0.3975953 0.27481738 0.6564321 #> #> -- bili (VI Rank: 2) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2356543 0.1536301 0.05872720 0.3719578 #> 0.80 0.2398021 0.1609720 0.06167673 0.3776136 #> 1.40 0.2613612 0.1809950 0.07893386 0.4064484 #> 3.52 0.3702763 0.3118827 0.17050712 0.5447088 #> 7.25 0.4780580 0.4406202 0.29442977 0.6434075 #> #> -- edema (VI Rank: 3) -------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3035731 0.1840849 0.06509174 0.5228237 #> 0.5 0.3558716 0.2649457 0.11132293 0.5831396 #> 1 0.4693915 0.3961470 0.28211662 0.6331870 #> #> -- copper (VI Rank: 4) ------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 25.5 0.2632768 0.1622871 0.05581251 0.4308234 #> 42.8 0.2707739 0.1703028 0.05887747 0.4418590 #> 74.0 0.2908707 0.1940176 0.07155433 0.4768302 #> 129 0.3444258 0.2651729 0.11918406 0.5574967 #> 214 0.4245218 0.3577346 0.21408331 0.6238041 #> #> Predicted risk at time t = 1826.25 for top 4 predictors head(as.data.table(pd_smry)) #> variable importance Value Mean Median 25th % 75th % #> 1: ascites 0.4960630 0 0.3083328 0.1985589 0.06581247 0.5241336 #> 2: ascites 0.4960630 1 0.4702396 0.3975953 0.27481738 0.6564321 #> 3: bili 0.4160074 0.60 0.2356543 0.1536301 0.05872720 0.3719578 #> 4: bili 0.4160074 0.80 0.2398021 0.1609720 0.06167673 0.3776136 #> 5: bili 0.4160074 1.40 0.2613612 0.1809950 0.07893386 0.4064484 #> 6: bili 0.4160074 3.52 0.3702763 0.3118827 0.17050712 0.5447088 #> pred_horizon level #> 1: 1826.25 0 #> 2: 1826.25 1 #> 3: 1826.25 #> 4: 1826.25 #> 5: 1826.25 #> 6: 1826.25 "},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"multiple-variables-jointly","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Multiple variables, jointly","title":"PD and ICE curves with ORSF","text":"Partial dependence can show expected value model’s predictions function specific predictor, function multiple predictors. instance, can estimate predicted risk joint function bili, edema, trt: inspection, model’s predictions indicate slightly lower risk placebo group, seem change much different values bili edema. clear increase predicted risk higher levels edema higher levels bili slope predicted risk function bili appears highest among patients edema 0.5. effect bili modified edema 0.5? quick sanity check coxph suggests .","code":"pred_spec = pred_spec_auto(bili, edema, trt) pd_bili_edema <- orsf_pd_oob(fit_surv, pred_spec) ggplot(pd_bili_edema) + aes(x = bili, y = medn, col = trt, linetype = edema) + geom_line() + labs(y = 'Expected predicted risk') library(survival) pbc_orsf$edema_05 <- ifelse(pbc_orsf$edema == '0.5', 'yes', 'no') fit_cph <- coxph(Surv(time,status) ~ edema_05 * bili, data = pbc_orsf) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> edema_05 -546.83 6.7248 1 0.009508 ** #> bili -513.59 66.4689 1 3.555e-16 *** #> edema_05:bili -510.54 6.1112 1 0.013433 * #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"find-interactions-using-pd","dir":"Articles","previous_headings":"Partial dependence (PD)","what":"Find interactions using PD","title":"PD and ICE curves with ORSF","text":"Random forests good using interactions, less good telling . Use orsf_vint() apply method variable interaction scoring PD described Greenwell et al (2018). can take little lots predictors. use sanity check coxph see interactions detected using standard test: Note: Caution warranted interpreting statistical hypotheses motivated data tested . Results like p-values interaction shown interpreted exploratory.","code":"vint_scores <- orsf_vint(fit_surv) vint_scores[1:5] #> interaction score #> 1: albumin..protime 1.1597307 #> 2: protime..stage 1.0710384 #> 3: copper..protime 0.7958742 #> 4: albumin..stage 0.7672757 #> 5: bili..stage 0.7619507 fit_cph <- coxph(Surv(time,status) ~ albumin * protime, data = pbc_orsf) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> albumin -526.29 47.801 1 4.717e-12 *** #> protime -514.89 22.806 1 1.792e-06 *** #> albumin:protime -511.76 6.252 1 0.01241 * #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 fit_cph <- update(fit_cph, . ~ stage * protime) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> stage -528.61 43.1626 3 2.273e-09 *** #> protime -516.05 25.1244 1 5.375e-07 *** #> stage:protime -514.06 3.9822 3 0.2634 #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 fit_cph <- update(fit_cph, . ~ copper * protime) anova(fit_cph) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> copper -524.72 50.9444 1 9.502e-13 *** #> protime -513.02 23.3872 1 1.325e-06 *** #> copper:protime -508.86 8.3216 1 0.003918 ** #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"individual-conditional-expectations-ice","dir":"Articles","previous_headings":"","what":"Individual conditional expectations (ICE)","title":"PD and ICE curves with ORSF","text":"Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"classification-1","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Classification","title":"PD and ICE curves with ORSF","text":"Compute ICE using --bag data flipper_length_mm = c(190, 210). two identifiers output: id_variable identifier current value variable(s) data. redundant one variable, helpful multiple variables. id_row identifier observation original data. Note predicted probabilities returned class observation data. Predicted probabilities given observation given variable value sum 1. example,","code":"pred_spec <- list(flipper_length_mm = c(190, 210)) ice_oob <- orsf_ice_oob(fit_clsf, pred_spec = pred_spec) ice_oob #> id_variable id_row class flipper_length_mm pred #> 1: 1 1 Adelie 190 0.92045213 #> 2: 1 2 Adelie 190 0.80427932 #> 3: 1 3 Adelie 190 0.84342550 #> 4: 1 4 Adelie 190 0.93514694 #> 5: 1 5 Adelie 190 0.97172229 #> --- #> 896: 2 146 Gentoo 210 0.25779089 #> 897: 2 147 Gentoo 210 0.04806888 #> 898: 2 148 Gentoo 210 0.07926342 #> 899: 2 149 Gentoo 210 0.84597108 #> 900: 2 150 Gentoo 210 0.10191162 ice_oob %>% .[flipper_length_mm == 190] %>% .[id_row == 1] %>% .[['pred']] %>% sum() #> [1] 1"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"regression-1","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Regression","title":"PD and ICE curves with ORSF","text":"Compute ICE using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"pred_spec <- list(flipper_length_mm = c(190, 210)) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row flipper_length_mm pred #> 1: 1 1 190 37.94483 #> 2: 1 2 190 37.61595 #> 3: 1 3 190 37.53681 #> 4: 1 4 190 39.49476 #> 5: 1 5 190 38.95635 #> --- #> 362: 2 179 210 51.80471 #> 363: 2 180 210 47.27183 #> 364: 2 181 210 47.05031 #> 365: 2 182 210 50.39028 #> 366: 2 183 210 48.44774 pred_spec = pred_spec_auto(species, island, body_mass_g) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row species island body_mass_g pred #> 1: 1 1 Adelie Biscoe 3200 37.78339 #> 2: 1 2 Adelie Biscoe 3200 37.73273 #> 3: 1 3 Adelie Biscoe 3200 37.71248 #> 4: 1 4 Adelie Biscoe 3200 40.25782 #> 5: 1 5 Adelie Biscoe 3200 40.04074 #> --- #> 8231: 45 179 Gentoo Torgersen 5300 46.14559 #> 8232: 45 180 Gentoo Torgersen 5300 43.98050 #> 8233: 45 181 Gentoo Torgersen 5300 44.59837 #> 8234: 45 182 Gentoo Torgersen 5300 44.85146 #> 8235: 45 183 Gentoo Torgersen 5300 44.23710 ice_new <- orsf_ice_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row variable value level pred #> 1: 1 1 species NA Adelie 37.74136 #> 2: 1 2 species NA Adelie 37.42367 #> 3: 1 3 species NA Adelie 37.04598 #> 4: 1 4 species NA Adelie 39.89602 #> 5: 1 5 species NA Adelie 39.14848 #> --- #> 2009: 5 179 body_mass_g 5300 51.50196 #> 2010: 5 180 body_mass_g 5300 47.27055 #> 2011: 5 181 body_mass_g 5300 48.34064 #> 2012: 5 182 body_mass_g 5300 48.75828 #> 2013: 5 183 body_mass_g 5300 48.11020 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') ice_new <- orsf_ice_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) ice_new #> id_variable id_row species island pred #> 1: 1 1 Adelie Biscoe 38.52327 #> 2: 1 2 Adelie Biscoe 38.32073 #> 3: 1 3 Adelie Biscoe 37.71248 #> 4: 1 4 Adelie Biscoe 41.68380 #> 5: 1 5 Adelie Biscoe 40.91140 #> --- #> 179: 1 179 Adelie Biscoe 43.09493 #> 180: 1 180 Adelie Biscoe 38.79455 #> 181: 1 181 Adelie Biscoe 39.37734 #> 182: 1 182 Adelie Biscoe 40.71952 #> 183: 1 183 Adelie Biscoe 39.34501"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"survival-1","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Survival","title":"PD and ICE curves with ORSF","text":"Compute ICE using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get ICE value: Multi-prediction horizon ice comes minimal extra computational cost. Use fine grid time values assess whether predictors time-varying effects.","code":"ice_train <- orsf_ice_inb(fit_surv, pred_spec = list(bili = 1:5)) ice_train #> id_variable id_row pred_horizon bili pred #> 1: 1 1 1826.25 1 0.9015162 #> 2: 1 2 1826.25 1 0.1019426 #> 3: 1 3 1826.25 1 0.6821646 #> 4: 1 4 1826.25 1 0.3623411 #> 5: 1 5 1826.25 1 0.1374271 #> --- #> 1376: 5 272 1826.25 5 0.2650957 #> 1377: 5 273 1826.25 5 0.3065318 #> 1378: 5 274 1826.25 5 0.3503776 #> 1379: 5 275 1826.25 5 0.1652897 #> 1380: 5 276 1826.25 5 0.3549165 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili)) ice_train #> id_variable id_row pred_horizon bili pred #> 1: 1 1 1826.25 0.60 0.89210440 #> 2: 1 2 1826.25 0.60 0.09186876 #> 3: 1 3 1826.25 0.60 0.65503431 #> 4: 1 4 1826.25 0.60 0.34622748 #> 5: 1 5 1826.25 0.60 0.13310425 #> --- #> 1376: 5 272 1826.25 7.25 0.31258148 #> 1377: 5 273 1826.25 7.25 0.35478676 #> 1378: 5 274 1826.25 7.25 0.41559176 #> 1379: 5 275 1826.25 7.25 0.25301890 #> 1380: 5 276 1826.25 7.25 0.44533769 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) ice_train #> id_variable id_row pred_horizon bili pred #> 1: 1 1 500 0.60 0.5949598 #> 2: 1 1 1000 0.60 0.7652137 #> 3: 1 1 1500 0.60 0.8751746 #> 4: 1 1 2000 0.60 0.9057135 #> 5: 1 1 2500 0.60 0.9231915 #> --- #> 8276: 5 276 1000 7.25 0.2111306 #> 8277: 5 276 1500 7.25 0.3642278 #> 8278: 5 276 2000 7.25 0.4850492 #> 8279: 5 276 2500 7.25 0.5720362 #> 8280: 5 276 3000 7.25 0.6206786"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"visualizing-ice-curves","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Visualizing ICE curves","title":"PD and ICE curves with ORSF","text":"Inspecting ICE curves observation can help identify whether heterogeneity model’s predictions. .e., effect variable follow pattern data, groups variable impacts risk differently? going turn boundary checking orsf_ice_oob setting boundary_checks = FALSE, allow generate ICE curves go beyond 90th percentile bili. plots, helpful scale ICE data. subtract initial value predicted risk (.e., bili = 1) observation’s conditional expectation values. , Every curve start 0 plot shows change predicted risk function bili. Now can visualize curves. inspection figure, individual slopes cluster around overall trend - Good! small number individual slopes appear flat. may helpful investigate .","code":"pred_spec <- list(bili = seq(1, 10, length.out = 25)) ice_oob <- orsf_ice_oob(fit_surv, pred_spec, boundary_checks = FALSE) ice_oob #> id_variable id_row pred_horizon bili pred #> 1: 1 1 1826.25 1 0.8790861 #> 2: 1 2 1826.25 1 0.8132035 #> 3: 1 3 1826.25 1 0.6240238 #> 4: 1 4 1826.25 1 0.7461603 #> 5: 1 5 1826.25 1 0.5754091 #> --- #> 6896: 25 272 1826.25 10 0.7018976 #> 6897: 25 273 1826.25 10 0.4606246 #> 6898: 25 274 1826.25 10 0.3347082 #> 6899: 25 275 1826.25 10 0.6046024 #> 6900: 25 276 1826.25 10 0.2789017 ice_oob[, pred_subtract := rep(pred[id_variable==1], times=25)] ice_oob[, pred := pred - pred_subtract] ggplot(ice_oob, aes(x = bili, y = pred, group = id_row)) + geom_line(alpha = 0.15) + labs(y = 'Change in predicted risk') + geom_smooth(se = FALSE, aes(group = 1))"},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"limitations-of-pd","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"Limitations of PD","title":"PD and ICE curves with ORSF","text":"Partial dependence number known limitations assumptions users aware (see Hooker, 2021). particular, partial dependence less intuitive >2 predictors examined jointly, assumed feature(s) partial dependence computed correlated features (likely true many cases). Accumulated local effect plots can used (see ) case feature independence valid assumption.","code":""},{"path":"https://bcjaeger.github.io/aorsf/articles/pd.html","id":"references","dir":"Articles","previous_headings":"Individual conditional expectations (ICE)","what":"References","title":"PD and ICE curves with ORSF","text":"Hooker, Giles, Mentch, Lucas, Zhou, Siyu (2021). “Unrestricted permutation forces extrapolation: variable importance requires least one model, free variable importance.” Statistics Computing, 31, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/authors.html","id":null,"dir":"","previous_headings":"","what":"Authors","title":"Authors and Citation","text":"Byron Jaeger. Author, maintainer. Nicholas Pajewski. Contributor. Sawyer Welden. Contributor. Christopher Jackson. Reviewer. Marvin Wright. Reviewer. Lukas Burk. Reviewer.","code":""},{"path":"https://bcjaeger.github.io/aorsf/authors.html","id":"citation","dir":"","previous_headings":"","what":"Citation","title":"Authors and Citation","text":"Jaeger et al. (2022). aorsf: R package supervised learning using oblique random survival forest. Journal Open Source Software, 7(77), 4705. https://doi.org/10.21105/joss.04705. Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM. Accelerated interpretable oblique random survival forests. Journal Computational Graphical Statistics. 2023 Aug 3:1-6. Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min YI, Mcclure LA, Howard G, Simon N. Oblique Random Survival Forests. Annals Applied Statistics. 13(3): 1847-1883. URL https://doi.org/10.1214/19-AOAS1261 DOI: 10.1214/19-AOAS1261","code":"@Article{, title = {aorsf: An R package for supervised learning using the oblique random survival forest}, author = {Byron C. Jaeger and Sawyer Welden and Kristin Lenoir and Nicholas M. Pajewski}, journal = {Journal of Open Source Software}, year = {2022}, volume = {7}, number = {77}, pages = {4705}, url = {https://doi.org/10.21105/joss.04705}, } @Article{, title = {Accelerated and interpretable oblique random survival forests}, author = {Byron C. Jaeger and Sawyer Welden and Kristin Lenoir and Jaime L. Speiser and Matthew W. Segar and Ambarish Pandey and Nicholas M. Pajewski}, journal = {Journal of Computational and Graphical Statistics}, year = {2023}, url = {https://doi.org/10.1080/10618600.2023.2231048}, } @Article{, title = {Oblique Random Survival Forests}, author = {Byron C. Jaeger and D. Leann Long and Dustin M. Long and Mario Sims and Jeff M. Szychowski and Yuan-I Min and Leslie A. Mcclure and George Howard and Noah Simon}, journal = {Annals of Applied Statistics}, year = {2019}, volume = {13}, number = {3}, pages = {1847--1883}, url = {https://doi.org/10.1214/19-AOAS1261}, }"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"aorsf-","dir":"","previous_headings":"","what":"Accelerated Oblique Random Forests","title":"Accelerated Oblique Random Forests","text":"Fit, interpret, make predictions oblique random forests (RFs).","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"why-aorsf","dir":"","previous_headings":"","what":"Why aorsf?","title":"Accelerated Oblique Random Forests","text":"Fast versatile tools oblique RFs.1 Accurate predictions.2 Intuitive design formula based interface. Extensive input checks informative error messages. Compatible tidymodels mlr3","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"installation","dir":"","previous_headings":"","what":"Installation","title":"Accelerated Oblique Random Forests","text":"can install aorsf CRAN using can install development version aorsf GitHub :","code":"install.packages(\"aorsf\") # install.packages(\"remotes\") remotes::install_github(\"ropensci/aorsf\")"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"get-started","dir":"","previous_headings":"","what":"Get started","title":"Accelerated Oblique Random Forests","text":"aorsf fits several types oblique RFs orsf() function, including classification, regression, survival RFs. classification, fit oblique RF predict penguin species using penguin data magnificent palmerpenguins R package regression, use data predict bill length penguins: personal favorite oblique survival RF accelerated Cox regression first type oblique RF aorsf provided (see JCGS paper). , use predict mortality risk following diagnosis primary biliary cirrhosis:","code":"library(aorsf) library(tidyverse) # An oblique classification RF penguin_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = species ~ .) penguin_fit #> ---------- Oblique random classification forest #> #> Linear combinations: Accelerated Logistic regression #> N observations: 333 #> N classes: 3 #> N trees: 5 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 6 #> Min observations in leaf: 5 #> OOB stat value: 0.99 #> OOB stat type: AUC-ROC #> Variable importance: anova #> #> ----------------------------------------- # An oblique regression RF bill_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = bill_length_mm ~ .) bill_fit #> ---------- Oblique random regression forest #> #> Linear combinations: Accelerated Linear regression #> N observations: 333 #> N trees: 5 #> N predictors total: 7 #> N predictors per node: 3 #> Average leaves per tree: 42.6 #> Min observations in leaf: 5 #> OOB stat value: 0.76 #> OOB stat type: RSQ #> Variable importance: anova #> #> ----------------------------------------- # An oblique survival RF pbc_fit <- orsf(data = pbc_orsf, n_tree = 5, formula = Surv(time, status) ~ . - id) pbc_fit #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 20.4 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.79 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"what-does-oblique-mean","dir":"","previous_headings":"","what":"What does “oblique” mean?","title":"Accelerated Oblique Random Forests","text":"Decision trees grown splitting set training data non-overlapping subsets, goal similarity within new subsets . subsets created single predictor, decision tree axis-based subset boundaries perpendicular axis predictor. linear combinations (.e., weighted sum) variables used instead single variable, tree oblique boundaries neither parallel perpendicular axis. Figure: Decision trees classification axis-based splitting (left) oblique splitting (right). Cases orange squares; controls purple circles. trees partition predictor space defined variables X1 X2, oblique splits better job separating two classes. , difference translate real data, impact random forests comprising hundreds axis-based oblique trees? demonstrate using penguin data.3 also use function make several plots: also use grid points plotting decision surfaces: use orsf mtry=1 fit axis-based trees: Next use orsf_update copy modify original model, expanding fit oblique tree using mtry=2 instead mtry=1, include 500 trees instead 1: now need visualize decision surfaces using predictions four fits: Figure: Axis-based oblique decision surfaces single tree ensemble 500 trees. Axis-based trees boundaries perpendicular predictor axes, whereas oblique trees can boundaries neither parallel perpendicular predictor axes. Axis-based forests tend ‘step-function’ decision boundaries, oblique forests tend smooth decision boundaries.","code":"plot_decision_surface <- function(predictions, title, grid){ # this is not a general function for plotting # decision surfaces. It just helps to minimize # copying and pasting of code. class_preds <- bind_cols(grid, predictions) %>% pivot_longer(cols = c(Adelie, Chinstrap, Gentoo)) %>% group_by(flipper_length_mm, bill_length_mm) %>% arrange(desc(value)) %>% slice(1) cols <- c(\"darkorange\", \"purple\", \"cyan4\") ggplot(class_preds, aes(bill_length_mm, flipper_length_mm)) + geom_contour_filled(aes(z = value, fill = name), alpha = .25) + geom_point(data = penguins_orsf, aes(color = species, shape = species), alpha = 0.5) + scale_color_manual(values = cols) + scale_fill_manual(values = cols) + labs(x = \"Bill length, mm\", y = \"Flipper length, mm\") + theme_minimal() + scale_x_continuous(expand = c(0,0)) + scale_y_continuous(expand = c(0,0)) + theme(panel.grid = element_blank(), panel.border = element_rect(fill = NA), legend.position = '') + labs(title = title) } grid <- expand_grid( flipper_length_mm = seq(min(penguins_orsf$flipper_length_mm), max(penguins_orsf$flipper_length_mm), len = 200), bill_length_mm = seq(min(penguins_orsf$bill_length_mm), max(penguins_orsf$bill_length_mm), len = 200) ) fit_axis_tree <- penguins_orsf %>% orsf(species ~ bill_length_mm + flipper_length_mm, n_tree = 1, mtry = 1, tree_seeds = 106760) fit_axis_forest <- fit_axis_tree %>% orsf_update(n_tree = 500) fit_oblique_tree <- fit_axis_tree %>% orsf_update(mtry = 2) fit_oblique_forest <- fit_oblique_tree %>% orsf_update(n_tree = 500) preds <- list(fit_axis_tree, fit_axis_forest, fit_oblique_tree, fit_oblique_forest) %>% map(predict, new_data = grid, pred_type = 'prob') titles <- c(\"Axis-based tree\", \"Axis-based forest\", \"Oblique tree\", \"Oblique forest\") plots <- map2(preds, titles, plot_decision_surface, grid = grid)"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"variable-importance","dir":"","previous_headings":"","what":"Variable importance","title":"Accelerated Oblique Random Forests","text":"importance individual predictor variables can estimated three ways using aorsf can used type oblique RF. Also, variable importance functions always return named character vector negation2: variable assessed separately multiplying variable’s coefficients -1 determining much model’s performance changes. worse model’s performance negating coefficients given variable, important variable. technique promising b/c require permutation emphasizes variables larger coefficients linear combinations, also relatively new hasn’t studied much permutation importance. See Jaeger, (2023) details technique. permutation: variable assessed separately randomly permuting variable’s values determining much model’s performance changes. worse model’s performance permuting values given variable, important variable. technique flexible, intuitive, frequently used. also several known limitations analysis variance (ANOVA)4: p-value computed coefficient linear combination variables decision tree. Importance individual predictor variable proportion times p-value coefficient < 0.01. technique efficient computationally, may effective permutation negation terms selecting signal noise variables. See Menze, 2011 details technique. can supply R function estimate --bag error (see oob vignette) estimate --bag variable importance (see orsf_vi examples)","code":"orsf_vi_negate(pbc_fit) #> bili copper stage sex age #> 0.1552460736 0.1156218837 0.0796917628 0.0533427094 0.0283132385 #> albumin trt chol alk.phos platelet #> 0.0279823814 0.0168238416 0.0153010749 0.0148718669 0.0094582765 #> edema ascites spiders protime hepato #> 0.0067975986 0.0065505801 0.0062356214 -0.0004653046 -0.0026664147 #> ast trig #> -0.0028902524 -0.0106616501 orsf_vi_permute(penguin_fit) #> bill_length_mm bill_depth_mm body_mass_g island #> 0.121351910 0.101846889 0.097822451 0.080772909 #> sex flipper_length_mm year #> 0.035053517 0.008270751 -0.008058339 orsf_vi_anova(bill_fit) #> species sex bill_depth_mm flipper_length_mm #> 0.51652893 0.27906977 0.06315789 0.04950495 #> body_mass_g island year #> 0.04807692 0.02687148 0.00000000"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"partial-dependence-pd","dir":"","previous_headings":"","what":"Partial dependence (PD)","title":"Accelerated Oblique Random Forests","text":"Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model’s prediction.. can use specific values predictor compute PD let aorsf pick reasonable values use pred_spec_auto(): summary function, orsf_summarize_uni(), computes PD many variables ask , using sensible values. PD, see vignette","code":"# pick your own values orsf_pd_oob(bill_fit, pred_spec = list(species = c(\"Adelie\", \"Gentoo\"))) #> species mean lwr medn upr #> #> 1: Adelie 39.99394 35.76532 39.80782 46.13931 #> 2: Gentoo 46.66565 40.02938 46.88517 51.61367 # let aorsf pick reasonable values for you: orsf_pd_oob(bill_fit, pred_spec = pred_spec_auto(bill_depth_mm, island)) #> bill_depth_mm island mean lwr medn upr #> #> 1: 14.3 Biscoe 43.94960 35.90421 45.30159 51.05109 #> 2: 15.6 Biscoe 44.24705 36.62759 45.57321 51.08020 #> 3: 17.3 Biscoe 44.84757 36.53804 45.62910 53.93833 #> 4: 18.7 Biscoe 45.08939 36.35893 46.16893 54.42075 #> 5: 19.5 Biscoe 45.13608 36.21033 46.08023 54.42075 #> --- #> 11: 14.3 Torgersen 43.55984 35.47143 44.18127 51.05109 #> 12: 15.6 Torgersen 43.77317 35.44683 44.28406 51.08020 #> 13: 17.3 Torgersen 44.56465 35.84585 44.83694 53.93833 #> 14: 18.7 Torgersen 44.68367 35.44010 44.86667 54.42075 #> 15: 19.5 Torgersen 44.64605 35.44010 44.86667 54.42075 orsf_summarize_uni(pbc_fit, n_variables = 2) #> #> -- bili (VI Rank: 1) ----------------------------- #> #> |----------------- Risk -----------------| #> Value Mean Median 25th % 75th % #> #> 0.60 0.2098108 0.07168855 0.01138461 0.2860450 #> 0.80 0.2117933 0.07692308 0.01709469 0.2884990 #> 1.40 0.2326560 0.08445419 0.02100837 0.3563622 #> 3.55 0.4265979 0.35820106 0.05128824 0.7342923 #> 7.30 0.4724608 0.44746241 0.11759259 0.8039683 #> #> -- copper (VI Rank: 2) --------------------------- #> #> |----------------- Risk -----------------| #> Value Mean Median 25th % 75th % #> #> 25.0 0.2332412 0.04425936 0.01587919 0.3888304 #> 42.5 0.2535448 0.07417582 0.01754386 0.4151786 #> 74.0 0.2825471 0.11111111 0.01988069 0.4770833 #> 130 0.3259604 0.18771003 0.04658385 0.5054348 #> 217 0.4213303 0.28571429 0.13345865 0.6859423 #> #> Predicted risk at time t = 1788 for top 2 predictors"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"individual-conditional-expectations-ice","dir":"","previous_headings":"","what":"Individual conditional expectations (ICE)","title":"Accelerated Oblique Random Forests","text":"Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor. ICE, see vignette","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"interaction-scores","dir":"","previous_headings":"","what":"Interaction scores","title":"Accelerated Oblique Random Forests","text":"orsf_vint() function computes score possible interaction model based PD using method described Greenwell et al, 2018.5 can slow larger datasets, substantial speedups occur making use multi-threading restricting search smaller set predictors. values score mean? values average standard deviation standard deviation PD one variable conditional variable. interpreted relative one another, .e., higher scoring interaction likely reflect real interaction two variables lower scoring one. interaction scores make sense? Let’s test top scoring lowest scoring interactions using coxph(). Note: exploratory true null hypothesis test. ? used data generate test null hypothesis. much conducting statistical inference test interactions coxph demonstrating interaction scores orsf_vint() provides consistent tests models.","code":"preds_interaction <- c(\"albumin\", \"protime\", \"bili\", \"spiders\", \"trt\") # While it is tempting to speed up `orsf_vint()` by growing a smaller # number of trees, results may become unstable with this shortcut. pbc_interactions <- pbc_fit %>% orsf_update(n_tree = 500, tree_seeds = 329) %>% orsf_vint(n_thread = 0, predictors = preds_interaction) pbc_interactions #> interaction score #> #> 1: albumin..protime 0.97837184 #> 2: protime..bili 0.78999788 #> 3: albumin..bili 0.59128756 #> 4: bili..spiders 0.13192184 #> 5: bili..trt 0.13192184 #> 6: albumin..spiders 0.06578222 #> 7: albumin..trt 0.06578222 #> 8: protime..spiders 0.03012718 #> 9: protime..trt 0.03012718 #> 10: spiders..trt 0.00000000 library(survival) # the top scoring interaction should get a lower p-value anova(coxph(Surv(time, status) ~ protime * albumin, data = pbc_orsf)) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> protime -538.51 23.353 1 1.349e-06 *** #> albumin -514.89 47.255 1 6.234e-12 *** #> protime:albumin -511.76 6.252 1 0.01241 * #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 # the bottom scoring interaction should get a higher p-value anova(coxph(Surv(time, status) ~ spiders * trt, data = pbc_orsf)) #> Analysis of Deviance Table #> Cox model: response is Surv(time, status) #> Terms added sequentially (first to last) #> #> loglik Chisq Df Pr(>|Chi|) #> NULL -550.19 #> spiders -538.58 23.2159 1 1.448e-06 *** #> trt -538.39 0.3877 1 0.5335 #> spiders:trt -538.29 0.2066 1 0.6494 #> --- #> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1"},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"comparison-to-existing-software","dir":"","previous_headings":"","what":"Comparison to existing software","title":"Accelerated Oblique Random Forests","text":"survival analysis, comparisons aorsf existing software presented JCGS paper. paper: describes aorsf detail summary procedures used tree fitting algorithm runs general benchmark comparing aorsf obliqueRSF several learners reports prediction accuracy computational efficiency learners. runs simulation study comparing variable importance techniques oblique survival RFs, axis based survival RFs, boosted trees. reports probability variable importance technique rank relevant variable higher importance irrelevant variable.","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"references","dir":"","previous_headings":"","what":"References","title":"Accelerated Oblique Random Forests","text":"Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min Y, Mcclure LA, Howard G, Simon N (2019). “Oblique random survival forests.” Annals Applied Statistics, 13(3). doi:10.1214/19-aoas1261 https://doi.org/10.1214/19-aoas1261. Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM (2023). “Accelerated interpretable oblique random survival forests.” Journal Computational Graphical Statistics, 1-16. doi:10.1080/10618600.2023.2231048 https://doi.org/10.1080/10618600.2023.2231048. Horst , Hill AP, Gorman KB (2020). palmerpenguins: Palmer Archipelago (Antarctica) penguin data. R package version 0.1.0, https://allisonhorst.github.io/palmerpenguins/. Menze, H B, Kelm, Michael B, Splitthoff, N D, Koethe, Ullrich, Hamprecht, F (2011). “oblique random forests.” Machine Learning Knowledge Discovery Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011, Proceedings, Part II 22, 453-469. Springer. Greenwell, M B, Boehmke, C B, McCarthy, J (2018). “simple effective model-based variable importance measure.” arXiv preprint arXiv:1805.04755.","code":""},{"path":"https://bcjaeger.github.io/aorsf/index.html","id":"funding","dir":"","previous_headings":"","what":"Funding","title":"Accelerated Oblique Random Forests","text":"developers aorsf received financial support Center Biomedical Informatics, Wake Forest University School Medicine. also received support National Center Advancing Translational Sciences National Institutes Health Award Number UL1TR001420. content solely responsibility authors necessarily represent official views National Institutes Health.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/aorsf-package.html","id":null,"dir":"Reference","previous_headings":"","what":"aorsf: Accelerated Oblique Random Forests — aorsf-package","title":"aorsf: Accelerated Oblique Random Forests — aorsf-package","text":"Fit, interpret, compute predictions oblique random forests. Includes support partial dependence, variable importance, passing customized functions variable importance identification linear combinations features. Methods oblique random survival forest described Jaeger et al., (2023) doi:10.1080/10618600.2023.2231048 .","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/aorsf-package.html","id":"author","dir":"Reference","previous_headings":"","what":"Author","title":"aorsf: Accelerated Oblique Random Forests — aorsf-package","text":"Maintainer: Byron Jaeger bjaeger@wakehealth.edu (ORCID) contributors: Nicholas Pajewski [contributor] Sawyer Welden swelden@wakehealth.edu [contributor] Christopher Jackson chris.jackson@mrc-bsu.cam.ac.uk [reviewer] Marvin Wright [reviewer] Lukas Burk [reviewer]","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":null,"dir":"Reference","previous_headings":"","what":"Coerce to data.table — as.data.table.orsf_summary_uni","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"Convert 'orsf_summary' object data.table object.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"","code":"# S3 method for orsf_summary_uni as.data.table(x, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"x object class 'orsf_summary_uni' ... used","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"data.table","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/as.data.table.orsf_summary_uni.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Coerce to data.table — as.data.table.orsf_summary_uni","text":"","code":"if (FALSE) { library(data.table) object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 25) smry <- orsf_summarize_uni(object, n_variables = 2) as.data.table(smry) }"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":null,"dir":"Reference","previous_headings":"","what":"Oblique Random Forests — orsf","title":"Oblique Random Forests — orsf","text":"Grow specify oblique random forest. name orsf() implies function works survival forests, can used classification, regression, survival forests.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Oblique Random Forests — orsf","text":"","code":"orsf( data, formula, control = NULL, weights = NULL, n_tree = 500, n_split = 5, n_retry = 3, n_thread = 0, mtry = NULL, sample_with_replacement = TRUE, sample_fraction = 0.632, leaf_min_events = 1, leaf_min_obs = 5, split_rule = NULL, split_min_events = 5, split_min_obs = 10, split_min_stat = NULL, oobag_pred_type = NULL, oobag_pred_horizon = NULL, oobag_eval_every = NULL, oobag_fun = NULL, importance = \"anova\", importance_max_pvalue = 0.01, group_factors = TRUE, tree_seeds = NULL, attach_data = TRUE, no_fit = FALSE, na_action = \"fail\", verbose_progress = FALSE, ... ) orsf_train(object, attach_data = TRUE)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Oblique Random Forests — orsf","text":"data data.frame, tibble, data.table contains relevant variables. formula (formula) Two sided formula single outcome. terms right names predictor variables, symbol '.' may used indicate variables data except response. symbol '-' may also used indicate removal predictor. Details response vary depending forest type: Classification: response single variable, variable type factor data. Regression: response single variable, variable typee double integer least 10 unique numeric values data. Survival: response include time variable, followed status variable, may written inside call Surv (see examples). control (orsf_control) object returned one orsf_control functions: orsf_control_survival, orsf_control_classification, orsf_control_regression. NULL (default) use accelerated control, fastest available option. survival classification, Cox Logistic regression 1 iteration, regression ordinary least squares. weights (numeric vector) Optional. given, input length equal nrow(data) complete imputed data length equal nrow(na.omit(data)) na_action \"omit\". weights vector used count observations events prior growing node tree, orsf() scales weights sum(weights) == nrow(data). helps make tree depth consistent weighted un-weighted fits. n_tree (integer) number trees grow. Default n_tree = 500. n_split (integer) number cut-points assessed splitting node decision trees. Default n_split = 5. n_retry (integer) node splittable, current linear combination inputs unable provide valid split, orsf try new linear combination based different set randomly selected predictors, n_retry times. Default n_retry = 3. Set n_retry = 0 prevent retries. n_thread (integer) number threads use growing trees, computing predictions, computing importance. Default 0, allows suitable number threads used based availability. mtry (integer) Number predictors randomly included candidates splitting node. default smallest integer greater square root number total predictors, .e., mtry = ceiling(sqrt(number predictors)) sample_with_replacement (logical) TRUE (default), observations sampled replacement -bag sample created decision tree. FALSE, observations sampled without replacement tree -bag sample containing sample_fraction% original sample. sample_fraction (double) proportion observations trees' -bag sample contain, relative number rows data. used sample_with_replacement FALSE. Default value 0.632. leaf_min_events (integer) input relevant survival analysis, specifies minimum number events leaf node. Default leaf_min_events = 1 leaf_min_obs (integer) minimum number observations leaf node. Default leaf_min_obs = 5. split_rule (character) assess quality potential splitting rule node. Valid options survival : 'logrank' : log-rank test statistic (default). 'cstat' : Harrell's concordance statistic. classification, valid options : 'gini' : gini impurity (default) 'cstat' : area underneath ROC curve (AUC-ROC) regression, valid options : 'variance' : variance reduction (default) split_min_events (integer) minimum number events required node consider splitting . Default split_min_events = 5. input relevant survival trees. split_min_obs (integer) minimum number observations required node consider splitting . Default split_min_obs = 10. split_min_stat (double) minimum test statistic required split node. splits found statistic exceeding split_min_stat, given node either becomes leaf retry occurs (n_retry retries). Defaults 3.84 split_rule = 'logrank' 0.55 split_rule = 'cstat' (see first note ) 0.00 split_rule = 'gini' (see second note ) 0.00 split_rule = 'variance' Note 1 C-statistic splitting, C < 0.50, consider statistic value 1 - C allow good 'anti-predictive' splits. , C-statistic initially computed 0.1, considered 1 - 0.10 = 0.90. Note 2 Gini impurity, value 0 1 usually indicate best worst possible scores, respectively. make things simple avoid introducing split_max_stat input, flip values Gini impurity 1 0 indicate best worst possible scores, respectively. oobag_pred_type (character) type --bag predictions compute fitting ensemble. Valid options tree type: 'none' : compute --bag predictions 'leaf' : ID predicted leaf returned tree Valid options survival: 'risk' : probability event occurring oobag_pred_horizon (default). 'surv' : 1 - risk. 'chf' : cumulative hazard function oobag_pred_horizon. 'mort' : mortality, .e., number events expected observations training data identical given observation. Valid options classification: 'prob' : probability class (default) 'class' : class (.e., .max(prob)) Valid options regression: 'mean' : mean value (default) oobag_pred_horizon (numeric) numeric value indicating time used --bag predictions. Default median observed times, .e., oobag_pred_horizon = median(time). input relevant survival trees prediction type 'risk', 'surv', 'chf'. oobag_eval_every (integer) --bag performance ensemble checked every oobag_eval_every trees. , oobag_eval_every = 10, --bag performance checked growing 10th tree, 20th tree, . Default oobag_eval_every = n_tree. oobag_fun (function) used evaluating --bag prediction accuracy every oobag_eval_every trees. oobag_fun = NULL (default), evaluation statistic selected based tree type survival: Harrell's C-statistic (1982) classification: Area underneath ROC curve (AUC-ROC) regression: Traditional prediction R-squared use oobag_fun note following: oobag_fun three inputs: y_mat, w_vec, s_vec survival trees, y_mat two column matrix first column named 'time' second named 'status'. classification trees, y_mat matrix number columns = number distinct classes outcome. regression, y_mat matrix one column. s_vec numeric vector containing predictions oobag_fun return numeric output length 1 details, see --bag vignette. importance (character) Indicate method variable importance: 'none': variable importance computed. 'anova': compute analysis variance (ANOVA) importance 'negate': compute negation importance 'permute': compute permutation importance details methods, see orsf_vi. importance_max_pvalue (double) relevant importance \"anova\". maximum p-value register positive case counting number times variable found 'significant' tree growth. Default 0.01, recommended Menze et al. group_factors (logical) relevant variable importance estimated. TRUE, importance factor variables reported overall aggregating importance individual levels factor. FALSE, importance individual factor levels returned. tree_seeds (integer vector) Optional. specified, random seeds set using values tree_seeds[] growing tree . Two forests grown number trees seeds exact --bag samples, making --bag error estimates forests comparable. NULL (default), seeds picked random. attach_data (logical) TRUE, copy training data attached output. required plan using functions like orsf_pd_oob orsf_summarize_uni interpret forest using training data. Default TRUE. no_fit (logical) TRUE, model fitting steps defined saved, training initiated. object returned can directly submitted orsf_train() long attach_data TRUE. na_action (character) happen data contains missing values (.e., NA values). Valid options : 'fail' : error thrown data contains NA values 'omit' : rows data incomplete data dropped 'impute_meanmode' : missing values continuous categorical variables data imputed using mean mode, respectively. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used). object untrained 'aorsf' object, created setting no_fit = TRUE orsf().","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Oblique Random Forests — orsf","text":"obliqueForest object","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Oblique Random Forests — orsf","text":"function called orf()? earlier versions, aorsf package exclusively oblique random survival forests. formula survival oblique RFs: response formula can survival object returned Surv function, can also just time status variables. .e., Surv(time, status) ~ . works time + status ~ . works response can also survival object stored data. example, y ~ . valid formula data$y inherits Surv class. mtry: mtry parameter may temporarily reduced ensure linear models used find combinations predictors remain stable. occurs coefficients linear model fitting algorithms may become infinite number predictors exceeds number observations. oobag_fun: oobag_fun specified, used compute negation importance permutation importance, role ANOVA importance. n_thread: R function called C++ (.e., user-supplied function compute --bag error identify linear combinations variables), n_thread automatically set 1 attempting run R functions multiple threads cause R session crash.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"what-is-an-oblique-decision-tree-","dir":"Reference","previous_headings":"","what":"What is an oblique decision tree?","title":"Oblique Random Forests — orsf","text":"Decision trees developed splitting set training data two new subsets, goal similarity within new subsets . splitting process repeated resulting subsets data stopping criterion met. new subsets data formed based single predictor, decision tree said axis-based splits data appear perpendicular axis predictor. linear combinations variables used instead single variable, tree oblique splits data neither parallel right angle axis Figure : Decision trees classification axis-based splitting (left) oblique splitting (right). Cases orange squares; controls purple circles. trees partition predictor space defined variables X1 X2, oblique splits better job separating two classes.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"what-is-a-random-forest-","dir":"Reference","previous_headings":"","what":"What is a random forest?","title":"Oblique Random Forests — orsf","text":"Random forests collections de-correlated decision trees. Predictions tree aggregated make ensemble prediction forest. details, see Breiman el, 2001.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"training-out-of-bag-error-and-testing","dir":"Reference","previous_headings":"","what":"Training, out-of-bag error, and testing","title":"Oblique Random Forests — orsf","text":"random forests, tree grown bootstrapped version training set. bootstrap samples selected replacement, bootstrapped training set contains two-thirds instances original training set. '--bag' data instances bootstrapped training set. tree random forest can make predictions --bag data, --bag predictions can aggregated make ensemble --bag prediction. Since --bag data used grow tree, accuracy ensemble --bag predictions approximate generalization error random forest. Generalization error refers error random forest's predictions applied predict outcomes data used train , .e., testing data.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Oblique Random Forests — orsf","text":"orsf() entry-point aorsf package. can used fit classification, regression, survival forests. classification, fit oblique RF predict penguin species using penguin data magnificent palmerpenguins R package regression, use data predict bill length penguins: personal favorite oblique survival RF accelerated Cox regression first type oblique RF aorsf provided (see ArXiv paper; paper also published Journal Computational Graphical Statistics publicly available ). , use predict mortality risk following diagnosis primary biliary cirrhosis:","code":"library(aorsf) library(magrittr) # for %>% # An oblique classification RF penguin_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = species ~ .) penguin_fit ## ---------- Oblique random classification forest ## ## Linear combinations: Accelerated Logistic regression ## N observations: 333 ## N classes: 3 ## N trees: 5 ## N predictors total: 7 ## N predictors per node: 3 ## Average leaves per tree: 4.4 ## Min observations in leaf: 5 ## OOB stat value: 0.99 ## OOB stat type: AUC-ROC ## Variable importance: anova ## ## ----------------------------------------- # An oblique regression RF bill_fit <- orsf(data = penguins_orsf, n_tree = 5, formula = bill_length_mm ~ .) bill_fit ## ---------- Oblique random regression forest ## ## Linear combinations: Accelerated Linear regression ## N observations: 333 ## N trees: 5 ## N predictors total: 7 ## N predictors per node: 3 ## Average leaves per tree: 51.2 ## Min observations in leaf: 5 ## OOB stat value: 0.77 ## OOB stat type: RSQ ## Variable importance: anova ## ## ----------------------------------------- # An oblique survival RF pbc_fit <- orsf(data = pbc_orsf, n_tree = 5, formula = Surv(time, status) ~ . - id) pbc_fit ## ---------- Oblique random survival forest ## ## Linear combinations: Accelerated Cox regression ## N observations: 276 ## N events: 111 ## N trees: 5 ## N predictors total: 17 ## N predictors per node: 5 ## Average leaves per tree: 20 ## Min observations in leaf: 5 ## Min events in leaf: 1 ## OOB stat value: 0.76 ## OOB stat type: Harrell's C-index ## Variable importance: anova ## ## -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"more-than-one-way-to-grow-a-forest","dir":"Reference","previous_headings":"","what":"More than one way to grow a forest","title":"Oblique Random Forests — orsf","text":"can use orsf(no_fit = TRUE) make specification grow forest instead fitted forest. ? Two reasons: computational tasks, may want check long take fit forest commit : fitting multiple forests, use blueprint along orsf_train() orsf_update() simplify code:","code":"orsf_spec <- orsf(pbc_orsf, formula = time + status ~ . - id, no_fit = TRUE) orsf_spec ## Untrained oblique random survival forest ## ## Linear combinations: Accelerated Cox regression ## N observations: 276 ## N events: 111 ## N trees: 500 ## N predictors total: 17 ## N predictors per node: 5 ## Average leaves per tree: 0 ## Min observations in leaf: 5 ## Min events in leaf: 1 ## OOB stat value: none ## OOB stat type: Harrell's C-index ## Variable importance: anova ## ## ----------------------------------------- orsf_spec %>% orsf_update(n_tree = 10000) %>% orsf_time_to_train() ## Time difference of 2.4331 secs orsf_fit <- orsf_train(orsf_spec) orsf_fit_10 <- orsf_update(orsf_fit, leaf_min_obs = 10) orsf_fit_20 <- orsf_update(orsf_fit, leaf_min_obs = 20) orsf_fit$leaf_min_obs ## [1] 5 orsf_fit_10$leaf_min_obs ## [1] 10 orsf_fit_20$leaf_min_obs ## [1] 20"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"tidymodels","dir":"Reference","previous_headings":"","what":"tidymodels","title":"Oblique Random Forests — orsf","text":"tidymodels includes support aorsf computational engine: Prediction aorsf models different times also supported:","code":"library(tidymodels) library(censored) library(yardstick) pbc_tidy <- pbc_orsf %>% mutate(event_time = Surv(time, status), .before = 1) %>% select(-c(id, time, status)) %>% as_tibble() split <- initial_split(pbc_tidy) orsf_spec <- rand_forest() %>% set_engine(\"aorsf\") %>% set_mode(\"censored regression\") orsf_fit <- fit(orsf_spec, formula = event_time ~ ., data = training(split)) time_points <- seq(500, 3000, by = 500) test_pred <- augment(orsf_fit, new_data = testing(split), eval_time = time_points) brier_scores <- test_pred %>% brier_survival(truth = event_time, .pred) brier_scores ## # A tibble: 6 x 4 ## .metric .estimator .eval_time .estimate ## ## 1 brier_survival standard 500 0.0466 ## 2 brier_survival standard 1000 0.0754 ## 3 brier_survival standard 1500 0.0612 ## 4 brier_survival standard 2000 0.0885 ## 5 brier_survival standard 2500 0.133 ## 6 brier_survival standard 3000 0.141 roc_scores <- test_pred %>% roc_auc_survival(truth = event_time, .pred) roc_scores ## # A tibble: 6 x 4 ## .metric .estimator .eval_time .estimate ## ## 1 roc_auc_survival standard 500 0.947 ## 2 roc_auc_survival standard 1000 0.939 ## 3 roc_auc_survival standard 1500 0.982 ## 4 roc_auc_survival standard 2000 0.961 ## 5 roc_auc_survival standard 2500 0.929 ## 6 roc_auc_survival standard 3000 0.942"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Oblique Random Forests — orsf","text":"Harrell, E F, Califf, M R, Pryor, B D, Lee, L K, Rosati, R (1982). \"Evaluating yield medical tests.\" Jama, 247(18), 2543-2546. Breiman, Leo (2001). \"Random Forests.\" Machine Learning, 45(1), 5-32. ISSN 1573-0565. Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS (2008). \"Random survival forests.\" Annals Applied Statistics, 2(3). Menze, H B, Kelm, Michael B, Splitthoff, N D, Koethe, Ullrich, Hamprecht, F (2011). \"oblique random forests.\" Machine Learning Knowledge Discovery Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011, Proceedings, Part II 22, 453-469. Springer. Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min Y, Mcclure LA, Howard G, Simon N (2019). \"Oblique random survival forests.\" Annals Applied Statistics, 13(3). Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM (2023). \"Accelerated interpretable oblique random survival forests.\" Journal Computational Graphical Statistics, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":null,"dir":"Reference","previous_headings":"","what":"Oblique random forest control — orsf_control","title":"Oblique random forest control — orsf_control","text":"Oblique random forest control","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Oblique random forest control — orsf_control","text":"","code":"orsf_control( tree_type, method, scale_x, ties, net_mix, target_df, max_iter, epsilon, ... ) orsf_control_classification( method = \"glm\", scale_x = TRUE, net_mix = 0.5, target_df = NULL, max_iter = 20, epsilon = 1e-09, ... ) orsf_control_regression( method = \"glm\", scale_x = TRUE, net_mix = 0.5, target_df = NULL, max_iter = 20, epsilon = 1e-09, ... ) orsf_control_survival( method = \"glm\", scale_x = TRUE, ties = \"efron\", net_mix = 0.5, target_df = NULL, max_iter = 20, epsilon = 1e-09, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Oblique random forest control — orsf_control","text":"tree_type (character) type tree. Valid options \"classification\", .e., categorical outcomes \"regression\", .e., continuous outcomes \"survival\", .e., time-event outcomes method (character function) identify linear linear combinations predictors. method character value, must one : 'glm': linear, logistic, cox regression 'net': 'glm' penalty terms 'pca': principal component analysis 'random': random draw uniform distribution method function, used identify linear combinations predictor variables. method must case accept three inputs named x_node, y_node w_node, expect following types dimensions: x_node (matrix; n rows, p columns) y_node (matrix; n rows, 2 columns) w_node (matrix; n rows, 1 column) addition, method must return matrix p rows 1 column. scale_x (logical) TRUE, values predictors scaled prior instance finding linear combination predictors, using summary values data current node decision tree. ties (character) character string specifying method tie handling. relevant modeling survival outcomes using method engages tied outcome times. ties, methods equivalent. Valid options 'breslow' 'efron'. Efron approximation default accurate dealing tied event times similar computational efficiency compared Breslow method. net_mix (double) elastic net mixing parameter. value 1 gives lasso penalty, value 0 gives ridge penalty. multiple values alpha given, penalized model fit using alpha value prior splitting node. target_df (integer) Preferred number variables used linear combination. example, mtry 5 target_df 3, sample 5 predictors look best linear combination using 3 . max_iter (integer) iteration continues convergence (see eps ) number attempted iterations equal iter_max. epsilon (double) using modeling based method, iteration continues algorithm relative change kind objective less epsilon, absolute change less sqrt(epsilon). ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Oblique random forest control — orsf_control","text":"object class 'orsf_control', used input control argument orsf. Components : tree_type: type trees fit lincomb_type: method linear combinations lincomb_eps: epsilon convergence lincomb_iter_max: max iterations lincomb_scale: scale . lincomb_alpha: mixing parameter lincomb_df_target: target degrees freedom lincomb_ties_method: method ties survival time lincomb_R_function: R function custom splits","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Oblique random forest control — orsf_control","text":"Adjust scale_x risk. Setting scale_x = FALSE reduce computation time also make orsf model dependent scale data, default value TRUE.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Oblique random forest control — orsf_control","text":"First load relevant packages","code":"set.seed(329730) suppressPackageStartupMessages({ library(aorsf) library(survival) library(ranger) library(riskRegression) })"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"accelerated-linear-combinations","dir":"Reference","previous_headings":"","what":"Accelerated linear combinations","title":"Oblique random forest control — orsf_control","text":"accelerated ORSF ensemble default nice balance computational speed prediction accuracy. runs single iteration Newton Raphson scoring Cox partial likelihood function find linear combinations predictors.","code":"fit_accel <- orsf(pbc_orsf, control = orsf_control_survival(), formula = Surv(time, status) ~ . - id, tree_seeds = 329)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"linear-combinations-with-cox-regression","dir":"Reference","previous_headings":"","what":"Linear combinations with Cox regression","title":"Oblique random forest control — orsf_control","text":"Setting inputs orsf_control_survival scale X matrix repeat iterations convergence allows run Cox regression non-terminal node survival tree, using regression coefficients create linear combinations predictors:","code":"control_cph <- orsf_control_survival(method = 'glm', scale_x = TRUE, max_iter = 20) fit_cph <- orsf(pbc_orsf, control = control_cph, formula = Surv(time, status) ~ . - id, tree_seeds = 329)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"linear-combinations-with-penalized-cox-regression","dir":"Reference","previous_headings":"","what":"Linear combinations with penalized cox regression","title":"Oblique random forest control — orsf_control","text":"Setting method == 'net' runs penalized Cox regression non-terminal node survival tree. can really helpful want feature selection within node, lot slower 'glm' option.","code":"# select 3 predictors out of 5 to be used in # each linear combination of predictors. control_net <- orsf_control_survival(method = 'net', target_df = 3) fit_net <- orsf(pbc_orsf, control = control_net, formula = Surv(time, status) ~ . - id, tree_seeds = 329)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control.html","id":"linear-combinations-with-your-own-function","dir":"Reference","previous_headings":"","what":"Linear combinations with your own function","title":"Oblique random forest control — orsf_control","text":"addition built-methods, customized functions can used identify linear combinations predictors. ’ll demonstrate . first uses random coefficients second derives coefficients principal component analysis third uses ranger() inside orsf(). approach similar method known reinforcement learning trees (see RLT package), although method “muting” crude compared method proposed Zhu et al.  can plug functions orsf_control_custom(), pass result orsf(): fit seems work best example? Let’s find evaluating --bag survival predictions. AUC values, highest lowest: indices prediction accuracy: inspection, net, accel, rlt high discrimination index prediction accuracy. rando pca less well, aren’t bad.","code":"f_rando <- function(x_node, y_node, w_node){ matrix(runif(ncol(x_node)), ncol=1) } f_pca <- function(x_node, y_node, w_node) { # estimate two principal components. pca <- stats::prcomp(x_node, rank. = 2) # use the second principal component to split the node pca$rotation[, 1L, drop = FALSE] } f_rlt <- function(x_node, y_node, w_node){ colnames(y_node) <- c('time', 'status') colnames(x_node) <- paste(\"x\", seq(ncol(x_node)), sep = '') data <- as.data.frame(cbind(y_node, x_node)) if(nrow(data) <= 10) return(matrix(runif(ncol(x_node)), ncol = 1)) fit <- ranger::ranger(data = data, formula = Surv(time, status) ~ ., num.trees = 25, num.threads = 1, min.node.size = 5, importance = 'permutation') out <- sort(fit$variable.importance, decreasing = TRUE) # \"mute\" the least two important variables n_vars <- length(out) if(n_vars > 4){ out[c(n_vars, n_vars-1)] <- 0 } # ensure out has same variable order as input out <- out[colnames(x_node)] # protect yourself out[is.na(out)] <- 0 matrix(out, ncol = 1) } fit_rando <- orsf(pbc_orsf, Surv(time, status) ~ . - id, control = orsf_control_survival(method = f_rando), tree_seeds = 329) fit_pca <- orsf(pbc_orsf, Surv(time, status) ~ . - id, control = orsf_control_survival(method = f_pca), tree_seeds = 329) fit_rlt <- orsf(pbc_orsf, time + status ~ . - id, control = orsf_control_survival(method = f_rlt), tree_seeds = 329) risk_preds <- list( accel = fit_accel$pred_oobag, cph = fit_cph$pred_oobag, net = fit_net$pred_oobag, rando = fit_rando$pred_oobag, pca = fit_pca$pred_oobag, rlt = fit_rlt$pred_oobag ) sc <- Score(object = risk_preds, formula = Surv(time, status) ~ 1, data = pbc_orsf, summary = 'IPA', times = fit_accel$pred_horizon) sc$AUC$score[order(-AUC)] ## model times AUC se lower upper ## 1: net 1788 0.9151649 0.02025057 0.8754745 0.9548553 ## 2: rlt 1788 0.9136390 0.02013475 0.8741756 0.9531024 ## 3: accel 1788 0.9095628 0.02143250 0.8675558 0.9515697 ## 4: cph 1788 0.9095628 0.02143250 0.8675558 0.9515697 ## 5: rando 1788 0.9062197 0.02148854 0.8641029 0.9483365 ## 6: pca 1788 0.8999479 0.02226683 0.8563057 0.9435901 sc$Brier$score[order(-IPA), .(model, times, IPA)] ## model times IPA ## 1: net 1788 0.4905777 ## 2: accel 1788 0.4806065 ## 3: cph 1788 0.4806065 ## 4: rlt 1788 0.4687322 ## 5: pca 1788 0.4383995 ## 6: rando 1788 0.4302814 ## 7: Null model 1788 0.0000000"},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":null,"dir":"Reference","previous_headings":"","what":"Cox regression ORSF control — orsf_control_cph","title":"Cox regression ORSF control — orsf_control_cph","text":"Use coefficients proportional hazards model create linear combinations predictor variables fitting orsf model.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Cox regression ORSF control — orsf_control_cph","text":"","code":"orsf_control_cph(method = \"efron\", eps = 1e-09, iter_max = 20, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Cox regression ORSF control — orsf_control_cph","text":"method (character) character string specifying method tie handling. ties, methods equivalent. Valid options 'breslow' 'efron'. Efron approximation default accurate dealing tied event times similar computational efficiency compared Breslow method. eps (double) using Newton Raphson scoring identify linear combinations inputs, iteration continues algorithm relative change log partial likelihood less eps, absolute change less sqrt(eps). Must positive. default value 1e-09 used consistency survival::coxph.control. iter_max (integer) iteration continues convergence (see eps ) number attempted iterations equal iter_max. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Cox regression ORSF control — orsf_control_cph","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Cox regression ORSF control — orsf_control_cph","text":"code survival package modified make routine. details Cox proportional hazards model, see coxph /Therneau Grambsch (2000).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_cph.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Cox regression ORSF control — orsf_control_cph","text":"Therneau T.M., Grambsch P.M. (2000) Cox Model. : Modeling Survival Data: Extending Cox Model. Statistics Biology Health. Springer, New York, NY. DOI: 10.1007/978-1-4757-3294-8_3","code":""},{"path":[]},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_custom.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Custom ORSF control — orsf_control_custom","text":"","code":"orsf_control_custom(beta_fun, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_custom.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Custom ORSF control — orsf_control_custom","text":"beta_fun (function) function define coefficients used linear combinations predictor variables. beta_fun must accept three inputs named x_node, y_node w_node, expect following types dimensions: x_node (matrix; n rows, p columns) y_node (matrix; n rows, 2 columns) w_node (matrix; n rows, 1 column) addition, beta_fun must return matrix p rows 1 column. conditions met, orsf_control_custom() let know. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_custom.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Custom ORSF control — orsf_control_custom","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":null,"dir":"Reference","previous_headings":"","what":"Accelerated ORSF control — orsf_control_fast","title":"Accelerated ORSF control — orsf_control_fast","text":"Fast methods identify linear combinations predictors fitting orsf model.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Accelerated ORSF control — orsf_control_fast","text":"","code":"orsf_control_fast(method = \"efron\", do_scale = TRUE, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Accelerated ORSF control — orsf_control_fast","text":"method (character) character string specifying method tie handling. ties, methods equivalent. Valid options 'breslow' 'efron'. Efron approximation default accurate dealing tied event times similar computational efficiency compared Breslow method. do_scale (logical) TRUE, values predictors scaled prior instance Newton Raphson scoring, using summary values data current node decision tree. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Accelerated ORSF control — orsf_control_fast","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_fast.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Accelerated ORSF control — orsf_control_fast","text":"code survival package modified make routine. Adjust do_scale risk. Setting do_scale = FALSE reduce computation time also make orsf model dependent scale data, default value TRUE.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":null,"dir":"Reference","previous_headings":"","what":"Penalized Cox regression ORSF control — orsf_control_net","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"Use regularized Cox proportional hazard models identify linear combinations input variables fitting orsf model.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"","code":"orsf_control_net(alpha = 1/2, df_target = NULL, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"alpha (double) elastic net mixing parameter. value 1 gives lasso penalty, value 0 gives ridge penalty. multiple values alpha given, penalized model fit using alpha value prior splitting node. df_target (integer) Preferred number variables used linear combination. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"object class 'orsf_control', used input control argument orsf.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"df_target less mtry, separate argument orsf indicates number variables chosen random prior finding linear combination variables.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_control_net.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Penalized Cox regression ORSF control — orsf_control_net","text":"Simon, Noah, Friedman, Jerome, Hastie, Trevor, Tibshirani, Rob (2011). \"Regularization paths Cox's proportional hazards model via coordinate descent.\" Journal statistical software, 39(5), 1.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":null,"dir":"Reference","previous_headings":"","what":"Individual Conditional Expectations — orsf_ice_oob","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Compute individual conditional expectations oblique random forest. Unlike partial dependence, shows expected prediction function one multiple predictors, individual conditional expectations (ICE) show prediction individual observation function predictor. can compute individual conditional expectations three ways using random forest: using -bag predictions training data using --bag predictions training data using predictions new set data See examples details","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Individual Conditional Expectations — orsf_ice_oob","text":"","code":"orsf_ice_oob( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_ice_inb( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_ice_new( object, pred_spec, new_data, pred_horizon = NULL, pred_type = NULL, na_action = \"fail\", expand_grid = TRUE, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Individual Conditional Expectations — orsf_ice_oob","text":"object (ObliqueForest) trained oblique random forest object (see orsf). pred_spec (named list, pspec_auto, data.frame). pred_spec named list, item list vector values used points partial dependence function. name item list indicate variable modified take corresponding values. pred_spec created using pred_spec_auto(), needed names variables use (see pred_spec_auto). pred_spec data.frame, columns indicate variable names, values indicate variable values, partial dependence computed using inputs row. pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_type (character) type predictions compute. Valid Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class regression: 'mean': predicted mean, .e., expected value expand_grid (logical) TRUE, partial dependence computed possible combinations inputs pred_spec. FALSE, partial dependence computed variable pred_spec, separately. boundary_checks (logical) TRUE, pred_spec checked make sure requested values 10th 90th percentile object's training data. FALSE, checks skipped. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used). new_data data.frame, tibble, data.table compute predictions . na_action (character) happen new_data contains missing values (.e., NA values). Valid options : 'fail' : error thrown new_data contains NA values 'omit' : rows new_data incomplete data dropped","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Individual Conditional Expectations — orsf_ice_oob","text":"data.table containing individual conditional expectations specified variable(s) , relevant, specified prediction horizon(s).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Individual Conditional Expectations — orsf_ice_oob","text":"can compute individual conditional expectation individual conditional expectations three ways: using -bag predictions training data. -bag individual conditional expectation indicates relationships model learned training. helpful goal interpret model. using --bag predictions training data. --bag individual conditional expectation indicates relationships model learned training using --bag data simulates application model new data. helpful want test model’s reliability fairness new data don’t access large testing set. using predictions new set data. New data individual conditional expectation shows model predicts outcomes observations seen. helpful want test model’s reliability fairness.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"classification","dir":"Reference","previous_headings":"","what":"Classification","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Begin fitting oblique classification random forest: Compute individual conditional expectation using --bag data flipper_length_mm = c(190, 210). two identifiers output: id_variable identifier current value variable(s) data. redundant one variable, helpful multiple variables. id_row identifier observation original data. Note predicted probabilities returned class observation data. Predicted probabilities given observation given variable value sum 1. example,","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) ice_oob <- orsf_ice_oob(fit_clsf, pred_spec = pred_spec) ice_oob ## id_variable id_row class flipper_length_mm pred ## 1: 1 1 Adelie 190 0.92059968 ## 2: 1 2 Adelie 190 0.80953569 ## 3: 1 3 Adelie 190 0.84869374 ## 4: 1 4 Adelie 190 0.93559660 ## 5: 1 5 Adelie 190 0.97708693 ## --- ## 896: 2 146 Gentoo 210 0.25636964 ## 897: 2 147 Gentoo 210 0.04798334 ## 898: 2 148 Gentoo 210 0.07945140 ## 899: 2 149 Gentoo 210 0.84811899 ## 900: 2 150 Gentoo 210 0.10695367 ice_oob %>% .[flipper_length_mm == 190] %>% .[id_row == 1] %>% .[['pred']] %>% sum() ## [1] 1"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"regression","dir":"Reference","previous_headings":"","what":"Regression","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Begin fitting oblique regression random forest: Compute individual conditional expectation using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row flipper_length_mm pred ## 1: 1 1 190 37.94483 ## 2: 1 2 190 37.61595 ## 3: 1 3 190 37.53681 ## 4: 1 4 190 39.49476 ## 5: 1 5 190 38.95635 ## --- ## 362: 2 179 210 51.80471 ## 363: 2 180 210 47.27183 ## 364: 2 181 210 47.05031 ## 365: 2 182 210 50.39028 ## 366: 2 183 210 48.44774 pred_spec = pred_spec_auto(species, island, body_mass_g) ice_new <- orsf_ice_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row species island body_mass_g pred ## 1: 1 1 Adelie Biscoe 3200 37.78339 ## 2: 1 2 Adelie Biscoe 3200 37.73273 ## 3: 1 3 Adelie Biscoe 3200 37.71248 ## 4: 1 4 Adelie Biscoe 3200 40.25782 ## 5: 1 5 Adelie Biscoe 3200 40.04074 ## --- ## 8231: 45 179 Gentoo Torgersen 5300 46.14559 ## 8232: 45 180 Gentoo Torgersen 5300 43.98050 ## 8233: 45 181 Gentoo Torgersen 5300 44.59837 ## 8234: 45 182 Gentoo Torgersen 5300 44.85146 ## 8235: 45 183 Gentoo Torgersen 5300 44.23710 ice_new <- orsf_ice_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row variable value level pred ## 1: 1 1 species NA Adelie 37.74136 ## 2: 1 2 species NA Adelie 37.42367 ## 3: 1 3 species NA Adelie 37.04598 ## 4: 1 4 species NA Adelie 39.89602 ## 5: 1 5 species NA Adelie 39.14848 ## --- ## 2009: 5 179 body_mass_g 5300 51.50196 ## 2010: 5 180 body_mass_g 5300 47.27055 ## 2011: 5 181 body_mass_g 5300 48.34064 ## 2012: 5 182 body_mass_g 5300 48.75828 ## 2013: 5 183 body_mass_g 5300 48.11020 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') ice_new <- orsf_ice_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) ice_new ## id_variable id_row species island pred ## 1: 1 1 Adelie Biscoe 38.52327 ## 2: 1 2 Adelie Biscoe 38.32073 ## 3: 1 3 Adelie Biscoe 37.71248 ## 4: 1 4 Adelie Biscoe 41.68380 ## 5: 1 5 Adelie Biscoe 40.91140 ## --- ## 179: 1 179 Adelie Biscoe 43.09493 ## 180: 1 180 Adelie Biscoe 38.79455 ## 181: 1 181 Adelie Biscoe 39.37734 ## 182: 1 182 Adelie Biscoe 40.71952 ## 183: 1 183 Adelie Biscoe 39.34501"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_ice_oob.html","id":"survival","dir":"Reference","previous_headings":"","what":"Survival","title":"Individual Conditional Expectations — orsf_ice_oob","text":"Begin fitting oblique survival random forest: Compute individual conditional expectation using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get individual conditional expectation value: Multi-prediction horizon ice comes minimal extra computational cost. Use fine grid time values assess whether predictors time-varying effects.","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) ice_train <- orsf_ice_inb(fit_surv, pred_spec = list(bili = 1:5)) ice_train ## id_variable id_row pred_horizon bili pred ## 1: 1 1 1826.25 1 0.1290317 ## 2: 1 2 1826.25 1 0.1242352 ## 3: 1 3 1826.25 1 0.0963452 ## 4: 1 4 1826.25 1 0.1172367 ## 5: 1 5 1826.25 1 0.2030256 ## --- ## 746: 5 146 1826.25 5 0.7868537 ## 747: 5 147 1826.25 5 0.2012954 ## 748: 5 148 1826.25 5 0.4893605 ## 749: 5 149 1826.25 5 0.4698220 ## 750: 5 150 1826.25 5 0.9557285 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili)) ice_train ## id_variable id_row pred_horizon bili pred ## 1: 1 1 1826.25 0.59 0.11706741 ## 2: 1 2 1826.25 0.59 0.11562173 ## 3: 1 3 1826.25 0.59 0.09110739 ## 4: 1 4 1826.25 0.59 0.10069721 ## 5: 1 5 1826.25 0.59 0.18769751 ## --- ## 746: 5 146 1826.25 7.21 0.82600898 ## 747: 5 147 1826.25 7.21 0.29156437 ## 748: 5 148 1826.25 7.21 0.58220919 ## 749: 5 149 1826.25 7.21 0.54168688 ## 750: 5 150 1826.25 7.21 0.96204106 ice_train <- orsf_ice_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) ice_train ## id_variable id_row pred_horizon bili pred ## 1: 1 1 500 0.59 0.008276627 ## 2: 1 1 1000 0.59 0.055715858 ## 3: 1 1 1500 0.59 0.084987224 ## 4: 1 1 2000 0.59 0.123090885 ## 5: 1 1 2500 0.59 0.165214938 ## --- ## 4496: 5 150 1000 7.21 0.835895969 ## 4497: 5 150 1500 7.21 0.932657591 ## 4498: 5 150 2000 7.21 0.965944498 ## 4499: 5 150 2500 7.21 0.970325309 ## 4500: 5 150 3000 7.21 0.979051377"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":null,"dir":"Reference","previous_headings":"","what":"Partial dependence — orsf_pd_oob","title":"Partial dependence — orsf_pd_oob","text":"Compute partial dependence oblique random forest. Partial dependence (PD) shows expected prediction model function single predictor multiple predictors. expectation marginalized values predictors, giving something like multivariable adjusted estimate model's prediction. can compute partial dependence three ways using random forest: using -bag predictions training data using --bag predictions training data using predictions new set data See examples details","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Partial dependence — orsf_pd_oob","text":"","code":"orsf_pd_oob( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, prob_values = c(0.025, 0.5, 0.975), prob_labels = c(\"lwr\", \"medn\", \"upr\"), boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_pd_inb( object, pred_spec, pred_horizon = NULL, pred_type = NULL, expand_grid = TRUE, prob_values = c(0.025, 0.5, 0.975), prob_labels = c(\"lwr\", \"medn\", \"upr\"), boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... ) orsf_pd_new( object, pred_spec, new_data, pred_horizon = NULL, pred_type = NULL, na_action = \"fail\", expand_grid = TRUE, prob_values = c(0.025, 0.5, 0.975), prob_labels = c(\"lwr\", \"medn\", \"upr\"), boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Partial dependence — orsf_pd_oob","text":"object (ObliqueForest) trained oblique random forest object (see orsf). pred_spec (named list, pspec_auto, data.frame). pred_spec named list, item list vector values used points partial dependence function. name item list indicate variable modified take corresponding values. pred_spec created using pred_spec_auto(), needed names variables use (see pred_spec_auto). pred_spec data.frame, columns indicate variable names, values indicate variable values, partial dependence computed using inputs row. pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_type (character) type predictions compute. Valid Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class regression: 'mean': predicted mean, .e., expected value expand_grid (logical) TRUE, partial dependence computed possible combinations inputs pred_spec. FALSE, partial dependence computed variable pred_spec, separately. prob_values (numeric) vector values 0 1, indicating quantiles used summarize partial dependence values set inputs. prob_values length prob_labels. quantiles calculated based predictions object set values indicated pred_spec. prob_labels (character) vector labels length prob_values, label indicating corresponding value prob_values labelled summarized outputs. prob_labels length prob_values. boundary_checks (logical) TRUE, pred_spec checked make sure requested values 10th 90th percentile object's training data. FALSE, checks skipped. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used). new_data data.frame, tibble, data.table compute predictions . na_action (character) happen new_data contains missing values (.e., NA values). Valid options : 'fail' : error thrown new_data contains NA values 'omit' : rows new_data incomplete data dropped","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Partial dependence — orsf_pd_oob","text":"data.table containing partial dependence values specified variable(s) , relevant, specified prediction horizon(s).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Partial dependence — orsf_pd_oob","text":"Partial dependence number known limitations assumptions users aware (see Hooker, 2021). particular, partial dependence less intuitive >2 predictors examined jointly, assumed feature(s) partial dependence computed correlated features (likely true many cases). Accumulated local effect plots can used (see ) case feature independence valid assumption.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Partial dependence — orsf_pd_oob","text":"can compute partial dependence individual conditional expectations three ways: using -bag predictions training data. -bag partial dependence indicates relationships model learned training. helpful goal interpret model. using --bag predictions training data. --bag partial dependence indicates relationships model learned training using --bag data simulates application model new data. helpful want test model’s reliability fairness new data don’t access large testing set. using predictions new set data. New data partial dependence shows model predicts outcomes observations seen. helpful want test model’s reliability fairness.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"classification","dir":"Reference","previous_headings":"","what":"Classification","title":"Partial dependence — orsf_pd_oob","text":"Begin fitting oblique classification random forest: Compute partial dependence using --bag data flipper_length_mm = c(190, 210). Note predicted probabilities returned class probabilities mean column sum 1 take sum class specific value pred_spec variables. example, isn’t case median predicted probability!","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_oob <- orsf_pd_oob(fit_clsf, pred_spec = pred_spec) pd_oob ## class flipper_length_mm mean lwr medn upr ## 1: Adelie 190 0.6180632 0.207463688 0.76047056 0.9809703 ## 2: Adelie 210 0.4346177 0.018583256 0.56486883 0.8647387 ## 3: Chinstrap 190 0.2119948 0.017692341 0.15658268 0.7163635 ## 4: Chinstrap 210 0.1801186 0.020454479 0.09525310 0.7085293 ## 5: Gentoo 190 0.1699420 0.001277844 0.02831331 0.5738689 ## 6: Gentoo 210 0.3852637 0.068685035 0.20853993 0.9537020 sum(pd_oob[flipper_length_mm == 190, mean]) ## [1] 1 sum(pd_oob[flipper_length_mm == 190, medn]) ## [1] 0.9453666"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"regression","dir":"Reference","previous_headings":"","what":"Regression","title":"Partial dependence — orsf_pd_oob","text":"Begin fitting oblique regression random forest: Compute partial dependence using new data flipper_length_mm = c(190, 210). can also let pred_spec_auto pick reasonable values like : default, combinations variables used. However, can also look variables one one, separately, like : can also bypass bells whistles using data.frame pred_spec. (Just make sure request values exist training data.)","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) pred_spec <- list(flipper_length_mm = c(190, 210)) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new ## flipper_length_mm mean lwr medn upr ## 1: 190 42.96571 37.09805 43.69769 48.72301 ## 2: 210 45.66012 40.50693 46.31577 51.65163 pred_spec = pred_spec_auto(species, island, body_mass_g) pd_new <- orsf_pd_new(fit_regr, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new ## species island body_mass_g mean lwr medn upr ## 1: Adelie Biscoe 3200 40.31374 37.24373 40.31967 44.22824 ## 2: Chinstrap Biscoe 3200 45.10582 42.63342 45.10859 47.60119 ## 3: Gentoo Biscoe 3200 42.81649 40.19221 42.55664 46.84035 ## 4: Adelie Dream 3200 40.16219 36.95895 40.34633 43.90681 ## 5: Chinstrap Dream 3200 46.21778 43.53954 45.90929 49.19173 ## 6: Gentoo Dream 3200 42.60465 39.89647 42.63520 46.28769 ## 7: Adelie Torgersen 3200 39.91652 36.80227 39.79806 43.68842 ## 8: Chinstrap Torgersen 3200 44.27807 41.95470 44.40742 46.68848 ## 9: Gentoo Torgersen 3200 42.09510 39.49863 41.80049 45.81833 ## 10: Adelie Biscoe 3550 40.77971 38.04027 40.59561 44.57505 ## 11: Chinstrap Biscoe 3550 45.81304 43.52102 45.73116 48.36366 ## 12: Gentoo Biscoe 3550 43.31233 40.77355 43.03077 47.22936 ## 13: Adelie Dream 3550 40.77741 38.07399 40.78175 44.37273 ## 14: Chinstrap Dream 3550 47.30926 44.80493 46.77540 50.47092 ## 15: Gentoo Dream 3550 43.26955 40.86119 43.16204 46.89190 ## 16: Adelie Torgersen 3550 40.25780 37.35251 40.07871 44.04576 ## 17: Chinstrap Torgersen 3550 44.77911 42.60161 44.81944 47.14986 ## 18: Gentoo Torgersen 3550 42.49520 39.95866 42.14160 46.26237 ## 19: Adelie Biscoe 3975 41.61744 38.94515 41.36634 45.38752 ## 20: Chinstrap Biscoe 3975 46.59363 44.59970 46.44923 49.11457 ## 21: Gentoo Biscoe 3975 44.07857 41.60792 43.74562 47.85109 ## 22: Adelie Dream 3975 41.50511 39.06187 41.24741 45.13027 ## 23: Chinstrap Dream 3975 48.14978 45.87390 47.54867 51.50683 ## 24: Gentoo Dream 3975 44.01928 41.70577 43.84099 47.50470 ## 25: Adelie Torgersen 3975 40.94764 38.12519 40.66759 44.73689 ## 26: Chinstrap Torgersen 3975 45.44820 43.49986 45.44036 47.63243 ## 27: Gentoo Torgersen 3975 43.13791 40.70628 42.70627 46.87306 ## 28: Adelie Biscoe 4700 42.93914 40.48463 42.44768 46.81756 ## 29: Chinstrap Biscoe 4700 47.18534 45.40866 47.07739 49.55747 ## 30: Gentoo Biscoe 4700 45.32541 43.08173 44.93498 49.23391 ## 31: Adelie Dream 4700 42.73806 40.44229 42.22226 46.49936 ## 32: Chinstrap Dream 4700 48.37354 46.34335 48.00781 51.18955 ## 33: Gentoo Dream 4700 45.09132 42.88328 44.79530 48.82180 ## 34: Adelie Torgersen 4700 42.09349 39.72074 41.56168 45.68838 ## 35: Chinstrap Torgersen 4700 46.17045 44.39042 46.09525 48.35127 ## 36: Gentoo Torgersen 4700 44.31621 42.18968 43.81773 47.98024 ## 37: Adelie Biscoe 5300 43.89769 41.43335 43.28504 48.10892 ## 38: Chinstrap Biscoe 5300 47.53721 45.66038 47.52770 49.88701 ## 39: Gentoo Biscoe 5300 46.16115 43.81722 45.59309 50.57469 ## 40: Adelie Dream 5300 43.59846 41.25825 43.24518 47.46193 ## 41: Chinstrap Dream 5300 48.48139 46.36282 48.25679 51.02996 ## 42: Gentoo Dream 5300 45.91819 43.62832 45.54110 49.91622 ## 43: Adelie Torgersen 5300 42.92879 40.66576 42.31072 46.76406 ## 44: Chinstrap Torgersen 5300 46.59576 44.80400 46.49196 49.03906 ## 45: Gentoo Torgersen 5300 45.11384 42.95190 44.51289 49.27629 ## species island body_mass_g mean lwr medn upr pd_new <- orsf_pd_new(fit_regr, expand_grid = FALSE, pred_spec = pred_spec, new_data = penguins_orsf_test) pd_new ## variable value level mean lwr medn upr ## 1: species NA Adelie 41.90271 37.10417 41.51723 48.51478 ## 2: species NA Chinstrap 47.11314 42.40419 46.96478 51.51392 ## 3: species NA Gentoo 44.37038 39.87306 43.89889 51.21635 ## 4: island NA Biscoe 44.21332 37.22711 45.27862 51.21635 ## 5: island NA Dream 44.43354 37.01471 45.57261 51.51392 ## 6: island NA Torgersen 43.29539 37.01513 44.26924 49.84391 ## 7: body_mass_g 3200 42.84625 37.03978 43.95991 49.19173 ## 8: body_mass_g 3550 43.53326 37.56730 44.43756 50.47092 ## 9: body_mass_g 3975 44.30431 38.31567 45.22089 51.50683 ## 10: body_mass_g 4700 45.22559 39.88199 46.34680 51.18955 ## 11: body_mass_g 5300 45.91412 40.84742 46.95327 51.48851 custom_pred_spec <- data.frame(species = 'Adelie', island = 'Biscoe') pd_new <- orsf_pd_new(fit_regr, pred_spec = custom_pred_spec, new_data = penguins_orsf_test) pd_new ## species island mean lwr medn upr ## 1: Adelie Biscoe 41.98024 37.22711 41.65252 48.51478"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"survival","dir":"Reference","previous_headings":"","what":"Survival","title":"Partial dependence — orsf_pd_oob","text":"Begin fitting oblique survival random forest: Compute partial dependence using -bag data bili = c(1,2,3,4,5): don’t specific values variable mind, let pred_spec_auto pick : Specify pred_horizon get partial dependence value: vector-valued pred_horizon input comes minimal extra computational cost. Use fine grid time values assess whether predictors time-varying effects. (see partial dependence vignette example)","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) pd_train <- orsf_pd_inb(fit_surv, pred_spec = list(bili = 1:5)) pd_train ## pred_horizon bili mean lwr medn upr ## 1: 1826.25 1 0.2566200 0.02234786 0.1334170 0.8918909 ## 2: 1826.25 2 0.3121392 0.06853733 0.1896849 0.9204338 ## 3: 1826.25 3 0.3703242 0.11409793 0.2578505 0.9416791 ## 4: 1826.25 4 0.4240692 0.15645214 0.3331057 0.9591581 ## 5: 1826.25 5 0.4663670 0.20123406 0.3841700 0.9655296 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili)) pd_train ## pred_horizon bili mean lwr medn upr ## 1: 1826.25 0.590 0.2484695 0.02035041 0.1243120 0.8823385 ## 2: 1826.25 0.725 0.2508045 0.02060111 0.1274237 0.8836536 ## 3: 1826.25 1.500 0.2797763 0.03964900 0.1601715 0.9041584 ## 4: 1826.25 3.500 0.3959349 0.13431288 0.2920400 0.9501230 ## 5: 1826.25 7.210 0.5344511 0.27869513 0.4651185 0.9782084 pd_train <- orsf_pd_inb(fit_surv, pred_spec_auto(bili), pred_horizon = seq(500, 3000, by = 500)) pd_train ## pred_horizon bili mean lwr medn upr ## 1: 500 0.590 0.06184375 0.0004433990 0.008765301 0.5918852 ## 2: 1000 0.590 0.14210619 0.0057937418 0.056124198 0.7381107 ## 3: 1500 0.590 0.20859307 0.0136094784 0.091808079 0.8577223 ## 4: 2000 0.590 0.26823465 0.0230476894 0.145707217 0.8918696 ## 5: 2500 0.590 0.31809404 0.0631155452 0.202189830 0.9035026 ## 6: 3000 0.590 0.39152139 0.0911566314 0.302738552 0.9239861 ## 7: 500 0.725 0.06255088 0.0004462367 0.008934806 0.5980510 ## 8: 1000 0.725 0.14337233 0.0063321712 0.056348007 0.7447805 ## 9: 1500 0.725 0.21058059 0.0140736894 0.093113771 0.8597396 ## 10: 2000 0.725 0.27056356 0.0235448705 0.146307939 0.8941464 ## 11: 2500 0.725 0.31922691 0.0626303822 0.202462648 0.9073970 ## 12: 3000 0.725 0.39426313 0.0911457406 0.308440546 0.9252028 ## 13: 500 1.500 0.06679162 0.0012717884 0.011028398 0.6241228 ## 14: 1000 1.500 0.15727919 0.0114789623 0.068332010 0.7678732 ## 15: 1500 1.500 0.23316655 0.0287320952 0.117289745 0.8789647 ## 16: 2000 1.500 0.30139227 0.0467927208 0.180096425 0.9144202 ## 17: 2500 1.500 0.35260943 0.0845866747 0.238015966 0.9266065 ## 18: 3000 1.500 0.43512074 0.1311103304 0.346025144 0.9438562 ## 19: 500 3.500 0.08638646 0.0052087533 0.028239001 0.6740930 ## 20: 1000 3.500 0.22353655 0.0519179775 0.139604845 0.8283986 ## 21: 1500 3.500 0.32700976 0.0901983241 0.217982772 0.9371150 ## 22: 2000 3.500 0.41618105 0.1445328597 0.311508093 0.9566091 ## 23: 2500 3.500 0.49248461 0.2195110942 0.402095677 0.9636221 ## 24: 3000 3.500 0.56008108 0.2635698957 0.503253258 0.9734948 ## 25: 500 7.210 0.12550962 0.0220920570 0.063425987 0.7526581 ## 26: 1000 7.210 0.32567558 0.1353851175 0.259047345 0.8875150 ## 27: 1500 7.210 0.46327019 0.2181840827 0.386681920 0.9700903 ## 28: 2000 7.210 0.55042753 0.2912654769 0.483477295 0.9812223 ## 29: 2500 7.210 0.61937483 0.3709845684 0.567895754 0.9844945 ## 30: 3000 7.210 0.67963922 0.4247511750 0.645083041 0.9888637 ## pred_horizon bili mean lwr medn upr"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_pd_oob.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Partial dependence — orsf_pd_oob","text":"Hooker, Giles, Mentch, Lucas, Zhou, Siyu (2021). \"Unrestricted permutation forces extrapolation: variable importance requires least one model, free variable importance.\" Statistics Computing, 31, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":null,"dir":"Reference","previous_headings":"","what":"Scale input data — orsf_scale_cph","title":"Scale input data — orsf_scale_cph","text":"functions exported users may access internal routines used scale inputs orsf_control_cph used.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Scale input data — orsf_scale_cph","text":"","code":"orsf_scale_cph(x_mat, w_vec = NULL) orsf_unscale_cph(x_mat)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Scale input data — orsf_scale_cph","text":"x_mat (numeric matrix) matrix values scaled unscaled. Note orsf_unscale_cph accept x_mat inputs attribute containing transform values, added automatically orsf_scale_cph. w_vec (numeric vector) optional vector weights. weights supplied (default), observations equally weighted. supplied, w_vec must length equal nrow(x_mat).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Scale input data — orsf_scale_cph","text":"scaled unscaled x_mat.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Scale input data — orsf_scale_cph","text":"data transformed first subtracting mean multiplying scale. inverse transform can completed using orsf_unscale_cph dividing column corresponding scale adding mean. values means scales stored attribute output returned orsf_scale_cph (see examples)","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_scale_cph.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Scale input data — orsf_scale_cph","text":"","code":"x_mat <- as.matrix(pbc_orsf[, c('bili', 'age', 'protime')]) head(x_mat) #> bili age protime #> 1 14.5 58.76523 12.2 #> 2 1.1 56.44627 10.6 #> 3 1.4 70.07255 12.0 #> 4 1.8 54.74059 10.3 #> 5 3.4 38.10541 10.9 #> 7 1.0 55.53457 9.7 x_scaled <- orsf_scale_cph(x_mat) head(x_scaled) #> bili age protime #> [1,] 3.77308887 1.0412574 1.9694656 #> [2,] -0.75476469 0.7719344 -0.1822316 #> [3,] -0.65339483 2.3544852 1.7005035 #> [4,] -0.51823502 0.5738373 -0.5856748 #> [5,] 0.02240421 -1.3581657 0.2212116 #> [6,] -0.78855464 0.6660494 -1.3925613 attributes(x_scaled) # note the transforms attribute #> $dim #> [1] 276 3 #> #> $dimnames #> $dimnames[[1]] #> NULL #> #> $dimnames[[2]] #> [1] \"bili\" \"age\" \"protime\" #> #> #> $transforms #> mean scale #> [1,] 3.333696 0.3378995 #> [2,] 49.799661 0.1161396 #> [3,] 10.735507 1.3448108 #> x_unscaled <- orsf_unscale_cph(x_scaled) head(x_unscaled) #> bili age protime #> [1,] 14.5 58.76523 12.2 #> [2,] 1.1 56.44627 10.6 #> [3,] 1.4 70.07255 12.0 #> [4,] 1.8 54.74059 10.3 #> [5,] 3.4 38.10541 10.9 #> [6,] 1.0 55.53457 9.7 # numeric difference in x_mat and x_unscaled should be practically 0 max(abs(x_mat - x_unscaled)) #> [1] 3.552714e-15"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":null,"dir":"Reference","previous_headings":"","what":"Univariate summary — orsf_summarize_uni","title":"Univariate summary — orsf_summarize_uni","text":"Summarize univariate information ORSF object","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Univariate summary — orsf_summarize_uni","text":"","code":"orsf_summarize_uni( object, n_variables = NULL, pred_horizon = NULL, pred_type = NULL, importance = NULL, verbose_progress = FALSE, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Univariate summary — orsf_summarize_uni","text":"object (ObliqueForest) trained oblique random forest object (see orsf). n_variables (integer) many variables summarized? Setting input lower number reduce computation time. pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_type (character) type predictions compute. Valid Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class regression: 'mean': predicted mean, .e., expected value importance (character) Indicate method variable importance: 'none': variable importance computed. 'anova': compute analysis variance (ANOVA) importance 'negate': compute negation importance 'permute': compute permutation importance details methods, see orsf_vi. verbose_progress (logical) TRUE, progress printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Univariate summary — orsf_summarize_uni","text":"object class 'orsf_summary', includes data importance individual predictors. expected values predictions specific values predictors.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Univariate summary — orsf_summarize_uni","text":"pred_horizon left unspecified, median value time--event variable object's training data used. recommended always specify prediction horizon, median time may especially meaningful horizon compute predicted risk values . object already variable importance values, can safely bypass computation variable importance function setting importance = 'none'.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_summarize_uni.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Univariate summary — orsf_summarize_uni","text":"","code":"object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 25) # since anova importance was used to make object, it is also # used for ranking variables in the summary, unless we specify # a different type of importance orsf_summarize_uni(object, n_variables = 2) #> #> -- bili (VI Rank: 1) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2313780 0.1479413 0.04781748 0.3280577 #> 0.80 0.2364004 0.1493642 0.05583256 0.3366922 #> 1.40 0.2650110 0.2064121 0.07340697 0.3652209 #> 3.52 0.3713478 0.3347867 0.17337079 0.5443745 #> 7.25 0.4685566 0.4452742 0.28893537 0.6351033 #> #> -- ascites (VI Rank: 2) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.2999651 0.2009324 0.06215675 0.5083512 #> 1 0.5279761 0.5040944 0.37091543 0.7086838 #> #> Predicted risk at time t = 1788 for top 2 predictors # if we want to summarize object according to variables # ranked by negation importance, we can compute negation # importance within orsf_summarize_uni() as follows: orsf_summarize_uni(object, n_variables = 2, importance = 'negate') #> #> -- bili (VI Rank: 1) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2313780 0.1479413 0.04781748 0.3280577 #> 0.80 0.2364004 0.1493642 0.05583256 0.3366922 #> 1.40 0.2650110 0.2064121 0.07340697 0.3652209 #> 3.52 0.3713478 0.3347867 0.17337079 0.5443745 #> 7.25 0.4685566 0.4452742 0.28893537 0.6351033 #> #> -- sex (VI Rank: 2) ---------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> m 0.3319983 0.2533052 0.11251192 0.5482338 #> f 0.3037761 0.1968855 0.06036558 0.5263657 #> #> Predicted risk at time t = 1788 for top 2 predictors"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":null,"dir":"Reference","previous_headings":"","what":"Estimate training time — orsf_time_to_train","title":"Estimate training time — orsf_time_to_train","text":"Estimate training time","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Estimate training time — orsf_time_to_train","text":"","code":"orsf_time_to_train(object, n_tree_subset = NULL)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Estimate training time — orsf_time_to_train","text":"object untrained aorsf object n_tree_subset (integer) many trees fit order estimate time needed train object. default value 10% trees specified object. .e., object n_tree 500, default value n_tree_subset 50.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Estimate training time — orsf_time_to_train","text":"difftime object.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_time_to_train.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Estimate training time — orsf_time_to_train","text":"","code":"# specify but do not train the model by setting no_fit = TRUE. object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 10, no_fit = TRUE) # approximate the time it will take to grow 10 trees time_estimated <- orsf_time_to_train(object, n_tree_subset=1) print(time_estimated) #> Time difference of 0.04523039 secs # let's see how close the approximation was time_true_start <- Sys.time() orsf_train(object) time_true_stop <- Sys.time() time_true <- time_true_stop - time_true_start print(time_true) #> Time difference of 0.0303793 secs # error abs(time_true - time_estimated) #> Time difference of 0.01485109 secs"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":null,"dir":"Reference","previous_headings":"","what":"Update Forest Parameters — orsf_update","title":"Update Forest Parameters — orsf_update","text":"Update Forest Parameters","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Update Forest Parameters — orsf_update","text":"","code":"orsf_update(object, ..., modify_in_place = FALSE, no_fit = NULL)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Update Forest Parameters — orsf_update","text":"object (ObliqueForest) oblique random forest object (see orsf). ... arguments plug orsf used define update. arguments include: data formula control weights n_tree n_split n_retry n_thread mtry sample_with_replacement sample_fraction leaf_min_events leaf_min_obs split_rule split_min_events split_min_obs split_min_stat pred_type oobag_pred_horizon oobag_eval_every oobag_fun importance importance_max_pvalue group_factors tree_seeds na_action verbose_progress Note can update control, change type forest. example, go classification regression orsf_update. modify_in_place (logical) TRUE, object modified inputs specified .... cautious, modification place overwrite existing data. FALSE (default), object copied modifications applied copy, leaving original object unmodified. no_fit (logical) TRUE, model fitting steps defined saved, training initiated. object returned can directly submitted orsf_train() long attach_data TRUE.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Update Forest Parameters — orsf_update","text":"ObliqueForest object.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Update Forest Parameters — orsf_update","text":"several dynamic inputs orsf default values NULL. Specifically, inputs control, weights, mtry, split_rule, split_min_stat, pred_type, pred_horizon, oobag_eval_function, tree_seeds, oobag_eval_every. explicit value given inputs call, re-formed. example, initial forest includes 17 predictors, default mtry smallest integer greater equal square root 17, .e., 5. , make updated forest 1 less predictor explicitly say mtry = 5, mtry re-initialized update based available 16 predictors, resulting value mtry 4. done avoid many potential errors occur dynamic outputs re-initialized.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_update.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Update Forest Parameters — orsf_update","text":"","code":"if (FALSE) { # initial fit has mtry of 5 fit <- orsf(pbc_orsf, time + status ~ . -id) # note that mtry is now 4 (see details) fit_new <- orsf_update(fit, formula = . ~ . - edema, n_tree = 100) # prevent dynamic updates by specifying inputs you want to freeze. fit_newer <- orsf_update(fit_new, mtry = 2) }"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":null,"dir":"Reference","previous_headings":"","what":"Variable Importance — orsf_vi","title":"Variable Importance — orsf_vi","text":"Estimate importance individual predictor variables using oblique random forests.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Variable Importance — orsf_vi","text":"","code":"orsf_vi( object, group_factors = TRUE, importance = NULL, oobag_fun = NULL, n_thread = NULL, verbose_progress = NULL, ... ) orsf_vi_negate( object, group_factors = TRUE, oobag_fun = NULL, n_thread = NULL, verbose_progress = NULL, ... ) orsf_vi_permute( object, group_factors = TRUE, oobag_fun = NULL, n_thread = NULL, verbose_progress = NULL, ... ) orsf_vi_anova(object, group_factors = TRUE, verbose_progress = NULL, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Variable Importance — orsf_vi","text":"object (ObliqueForest) trained oblique random forest object (see orsf). group_factors (logical) TRUE, importance factor variables reported overall aggregating importance individual levels factor. FALSE, importance individual factor levels returned. importance (character) Indicate method variable importance: 'anova': compute analysis variance (ANOVA) importance 'negate': compute negation importance 'permute': compute permutation importance oobag_fun (function) used evaluating --bag prediction accuracy negating coefficients (importance = 'negate') permuting values predictor (importance = 'permute') oobag_fun = NULL (default), evaluation statistic selected based tree type survival: Harrell's C-statistic (1982) classification: Area underneath ROC curve (AUC-ROC) regression: Traditional prediction R-squared use oobag_fun note following: oobag_fun three inputs: y_mat, w_vec, s_vec survival trees, y_mat two column matrix first column named 'time' second named 'status'. classification trees, y_mat matrix number columns = number distinct classes outcome. regression, y_mat matrix one column. s_vec numeric vector containing predictions oobag_fun return numeric output length 1 oobag_fun used created object initial value --bag prediction accuracy consistent values computed variable importance estimated. details, see --bag vignette. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Variable Importance — orsf_vi","text":"orsf_vi functions return named numeric vector. Names vector predictor variables used object Values vector estimated importance given predictor. returned vector sorted highest lowest value, higher values indicating higher importance.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Variable Importance — orsf_vi","text":"ObliqueForest object grown importance = 'anova', 'negate', 'permute', output vector importance values based requested type importance. However, orsf_vi() can used compute variable importance growing forest compute different type importance. orsf_vi() general purpose function extract compute variable importance estimates ObliqueForest object (see orsf). orsf_vi_negate(), orsf_vi_permute(), orsf_vi_anova() wrappers orsf_vi(). way functions work depends whether object given already variable importance estimates (see examples).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"variable-importance-methods","dir":"Reference","previous_headings":"","what":"Variable importance methods","title":"Variable Importance — orsf_vi","text":"negation importance: variable assessed separately multiplying variable's coefficients -1 determining much model's performance changes. worse model's performance negating coefficients given variable, important variable. technique promising b/c require permutation emphasizes variables larger coefficients linear combinations, also relatively new studied much permutation importance. See Jaeger, (2023) details technique. permutation importance: variable assessed separately randomly permuting variable's values determining much model's performance changes. worse model's performance permuting values given variable, important variable. technique flexible, intuitive, frequently used. also several known limitations analysis variance (ANOVA) importance: p-value computed coefficient linear combination variables decision tree. Importance individual predictor variable proportion times p-value coefficient < 0.01. technique efficient computationally, may effective permutation negation terms selecting signal noise variables. See Menze, 2011 details technique.","code":""},{"path":[]},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"anova-importance","dir":"Reference","previous_headings":"","what":"ANOVA importance","title":"Variable Importance — orsf_vi","text":"default variable importance technique, ANOVA, calculated fit oblique random forest ensemble. ANOVA default fast, may decisive permutation negation techniques variable selection.","code":"fit <- orsf(pbc_orsf, Surv(time, status) ~ . - id) fit ## ---------- Oblique random survival forest ## ## Linear combinations: Accelerated Cox regression ## N observations: 276 ## N events: 111 ## N trees: 500 ## N predictors total: 17 ## N predictors per node: 5 ## Average leaves per tree: 21.022 ## Min observations in leaf: 5 ## Min events in leaf: 1 ## OOB stat value: 0.84 ## OOB stat type: Harrell's C-index ## Variable importance: anova ## ## -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"raw-vi-values","dir":"Reference","previous_headings":"","what":"Raw VI values","title":"Variable Importance — orsf_vi","text":"‘raw’ variable importance values can accessed fit object ‘raw’ values factors aggregated single value. Currently one value k-1 levels k level factor. example, can see edema_1 edema_0.5 importance values edema factor variable levels 0, 0.5, 1.","code":"fit$get_importance_raw() ## [,1] ## trt_placebo 0.06355042 ## age 0.23259259 ## sex_f 0.14700432 ## ascites_1 0.46791708 ## hepato_1 0.14349776 ## spiders_1 0.17371938 ## edema_0.5 0.17459191 ## edema_1 0.51197605 ## bili 0.40590758 ## chol 0.17666667 ## albumin 0.25972156 ## copper 0.28840580 ## alk.phos 0.10614251 ## ast 0.18327491 ## trig 0.12815626 ## platelet 0.09265648 ## protime 0.22656250 ## stage 0.20264766"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"collapse-vi-across-factor-levels","dir":"Reference","previous_headings":"","what":"Collapse VI across factor levels","title":"Variable Importance — orsf_vi","text":"get aggregated values across levels factor, access importance element orsf fit: use orsf_vi() group_factors set TRUE (default) Note can make default returned importance values ungrouped setting group_factors FALSE orsf_vi functions orsf function.","code":"# this assumes you used group_factors = TRUE in orsf() fit$importance ## ascites bili edema copper albumin age protime ## 0.46791708 0.40590758 0.31115216 0.28840580 0.25972156 0.23259259 0.22656250 ## stage ast chol spiders sex hepato trig ## 0.20264766 0.18327491 0.17666667 0.17371938 0.14700432 0.14349776 0.12815626 ## alk.phos platelet trt ## 0.10614251 0.09265648 0.06355042 orsf_vi(fit) ## ascites bili edema copper albumin age protime ## 0.46791708 0.40590758 0.31115216 0.28840580 0.25972156 0.23259259 0.22656250 ## stage ast chol spiders sex hepato trig ## 0.20264766 0.18327491 0.17666667 0.17371938 0.14700432 0.14349776 0.12815626 ## alk.phos platelet trt ## 0.10614251 0.09265648 0.06355042"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"add-vi-to-an-oblique-random-forest","dir":"Reference","previous_headings":"","what":"Add VI to an oblique random forest","title":"Variable Importance — orsf_vi","text":"can fit oblique random forest without VI, add VI later","code":"fit_no_vi <- orsf(pbc_orsf, Surv(time, status) ~ . - id, importance = 'none') # Note: you can't call orsf_vi_anova() on fit_no_vi because anova # VI can only be computed while the forest is being grown. orsf_vi_negate(fit_no_vi) ## bili copper sex protime age stage ## 0.130439814 0.051880867 0.038308025 0.025115249 0.023826061 0.020354822 ## albumin ascites chol ast spiders hepato ## 0.019997729 0.015918292 0.013320469 0.010086726 0.007409116 0.007326714 ## edema trt alk.phos trig platelet ## 0.006844435 0.003214544 0.002517057 0.002469545 0.001056829 orsf_vi_permute(fit_no_vi) ## bili copper age ascites protime ## 0.0592069141 0.0237362075 0.0136479213 0.0130805894 0.0123091354 ## stage albumin chol hepato ast ## 0.0117177661 0.0106414724 0.0064501213 0.0058813969 0.0057753740 ## edema spiders sex trig platelet ## 0.0052171180 0.0048427005 0.0023386947 0.0017883700 0.0013533691 ## alk.phos trt ## 0.0006492029 -0.0009921507"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"oblique-random-forest-and-vi-all-at-once","dir":"Reference","previous_headings":"","what":"Oblique random forest and VI all at once","title":"Variable Importance — orsf_vi","text":"fit oblique random forest compute vi time can still get negation VI fit, needs computed","code":"fit_permute_vi <- orsf(pbc_orsf, Surv(time, status) ~ . - id, importance = 'permute') # get the vi instantly (i.e., it doesn't need to be computed again) orsf_vi_permute(fit_permute_vi) ## bili copper ascites protime albumin ## 0.0571305446 0.0243657146 0.0138318057 0.0133401675 0.0130746154 ## age stage chol ast spiders ## 0.0123610374 0.0102963203 0.0077895394 0.0075250059 0.0048628813 ## edema hepato sex platelet trig ## 0.0046003168 0.0039818730 0.0016891584 0.0012767063 0.0007324402 ## alk.phos trt ## 0.0005128897 -0.0014443967 orsf_vi_negate(fit_permute_vi) ## bili copper sex protime stage age ## 0.123331760 0.052544318 0.037291358 0.024977898 0.023239189 0.021934511 ## albumin ascites chol ast spiders edema ## 0.020586632 0.014229536 0.014053040 0.012227048 0.007643156 0.006832766 ## hepato trt alk.phos trig platelet ## 0.006301693 0.004348705 0.002371797 0.002309396 0.001347035"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"custom-functions-for-vi","dir":"Reference","previous_headings":"","what":"Custom functions for VI","title":"Variable Importance — orsf_vi","text":"default prediction accuracy functions work well time: sometimes want something specific defaults just won’t work. cases, can compute VI function ’d like measure prediction accuracy supplying valid function oobag_fun input. example, use mean absolute error . Higher values considered good aorsf computes prediction accuracy, make function return pseudo R-squared based mean absolute error:","code":"fit_standard <- orsf(penguins_orsf, bill_length_mm ~ ., tree_seeds = 1) # Default method for prediction accuracy with VI is R-squared orsf_vi_permute(fit_standard) ## species flipper_length_mm body_mass_g bill_depth_mm ## 0.3725898166 0.3261834607 0.2225730676 0.1026569498 ## island sex year ## 0.0876071687 0.0844807334 0.0006978493 rsq_mae <- function(y_mat, w_vec, s_vec){ mae_standard <- mean(abs((y_mat - mean(y_mat)) * w_vec)) mae_fit <- mean(abs((y_mat - s_vec) * w_vec)) 1 - mae_fit / mae_standard } fit_custom <- orsf_update(fit_standard, oobag_fun = rsq_mae) # not much changes, but the difference between variables shrinks # and the ordering of sex and island has swapped orsf_vi_permute(fit_custom) ## species flipper_length_mm body_mass_g bill_depth_mm ## 0.206951751 0.193248912 0.140899603 0.076759148 ## sex island year ## 0.073042331 0.050851073 0.003633365"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vi.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Variable Importance — orsf_vi","text":"Harrell, E F, Califf, M R, Pryor, B D, Lee, L K, Rosati, R (1982). \"Evaluating yield medical tests.\" Jama, 247(18), 2543-2546. Breiman, Leo (2001). \"Random Forests.\" Machine Learning, 45(1), 5-32. ISSN 1573-0565. Menze, H B, Kelm, Michael B, Splitthoff, N D, Koethe, Ullrich, Hamprecht, F (2011). \"oblique random forests.\" Machine Learning Knowledge Discovery Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011, Proceedings, Part II 22, 453-469. Springer. Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey , Pajewski NM (2023). \"Accelerated interpretable oblique random survival forests.\" Journal Computational Graphical Statistics, 1-16.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":null,"dir":"Reference","previous_headings":"","what":"Variable Interactions — orsf_vint","title":"Variable Interactions — orsf_vint","text":"Use variable interaction score described Greenwell et al (2018). method can computationally demanding, using n_thread=0 can substantially reduce time needed compute scores.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Variable Interactions — orsf_vint","text":"","code":"orsf_vint( object, predictors = NULL, n_thread = NULL, verbose_progress = NULL, sep = \"..\" )"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Variable Interactions — orsf_vint","text":"object (ObliqueForest) trained oblique random forest object (see orsf) predictors (character) vector length 2 names predictors used object. pairwise interactions predictors scored. NULL (default), predictors used. n_thread (integer) number threads use growing trees, computing predictions, computing importance. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. sep (character) separate names two predictors. default value \"..\" returns names name1..name2","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Variable Interactions — orsf_vint","text":"data.table variable interaction scores","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Variable Interactions — orsf_vint","text":"number possible interactions grows exponentially based number predictors. caution warranted using large predictor sets recommended supply specific vector predictor names assess rather global search. good strategy use n_tree = 5 search predictors, pick top 10 interactions, get unique predictors , re-run just predictors trees.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Variable Interactions — orsf_vint","text":"Greenwell, M B, Boehmke, C B, McCarthy, J (2018). \"simple effective model-based variable importance measure.\" arXiv preprint arXiv:1805.04755.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vint.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Variable Interactions — orsf_vint","text":"","code":"set.seed(329) data <- data.frame( x1 = rnorm(500), x2 = rnorm(500), x3 = rnorm(500) ) data$y = with(data, expr = x1 + x2 + x3 + 1/2*x1 * x2 + x2 * x3 + rnorm(500)) forest <- orsf(data, y ~ ., n_tree = 5) orsf_vint(forest) #> interaction score #> 1: x2..x3 0.8021932 #> 2: x1..x2 0.5095065 #> 3: x1..x3 0.1133252"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":null,"dir":"Reference","previous_headings":"","what":"Variable selection — orsf_vs","title":"Variable selection — orsf_vs","text":"Variable selection","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Variable selection — orsf_vs","text":"","code":"orsf_vs(object, n_predictor_min = 3, verbose_progress = NULL)"},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Variable selection — orsf_vs","text":"object (ObliqueForest) trained oblique random forest object (see orsf). n_predictor_min (integer) minimum number predictors allowed verbose_progress (logical) implemented yet. progress printed console?","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Variable selection — orsf_vs","text":"data.table four columns: n_predictors: number predictors used stat_value: --bag statistic predictors_included: names predictors included predictor_dropped: predictor selected dropped","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Variable selection — orsf_vs","text":"tree_seeds specified object successive run orsf evaluated --bag samples initial run.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/orsf_vs.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Variable selection — orsf_vs","text":"","code":"object <- orsf(formula = time + status ~ ., data = pbc_orsf, n_tree = 25, importance = 'anova') orsf_vs(object, n_predictor_min = 15) #> n_predictors stat_value predictors_included #> 1: 15 0.8356685 id,age,sex_f,ascites_1,spiders_1,edema_0.5,... #> 2: 16 0.8351997 id,age,sex_f,ascites_1,hepato_1,spiders_1,... #> 3: 17 0.8296786 id,trt_placebo,age,sex_f,ascites_1,hepato_1,... #> 4: 18 0.8185322 id,trt_placebo,age,sex_f,ascites_1,hepato_1,... #> predictor_dropped #> 1: platelet #> 2: hepato_1 #> 3: trt_placebo #> 4: alk.phos"},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":null,"dir":"Reference","previous_headings":"","what":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"data light modification survival::pbc data. modifications :","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"","code":"pbc_orsf"},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"data frame 276 rows 20 variables: id case number time number days registration earlier death, transplantion, study analysis July, 1986 status status endpoint, 0 censored transplant, 1 dead trt randomized treatment group: D-penicillmain placebo age years sex m/f ascites presence ascites hepato presence hepatomegaly enlarged liver spiders blood vessel malformations skin edema 0 edema, 0.5 untreated successfully treated, 1 edema despite diuretic therapy bili serum bilirubin (mg/dl) chol serum cholesterol (mg/dl) albumin serum albumin (g/dl) copper urine copper (ug/day) alk.phos alkaline phosphotase (U/liter) ast aspartate aminotransferase, called SGOT (U/ml) trig triglycerides (mg/dl) platelet platelet count protime standardized blood clotting time stage histologic stage disease (needs biopsy)","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"T Therneau P Grambsch (2000), Modeling Survival Data: Extending Cox Model, Springer-Verlag, New York. ISBN: 0-387-98784-3.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pbc_orsf.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Mayo Clinic Primary Biliary Cholangitis Data — pbc_orsf","text":"removed rows missing data converted status 0 censor transplant, 1 dead converted stage ordered factor. converted trt, ascites, hepato, spiders, edema factors.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":null,"dir":"Reference","previous_headings":"","what":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"data copied lightly modified penguins data palmerpenguins R package. modification removal rows missing data. data include measurements penguin species, island Palmer Archipelago, size (flipper length, body mass, bill dimensions), sex.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"","code":"penguins_orsf"},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"tibble 333 rows 8 variables: species factor denoting penguin species (Adélie, Chinstrap Gentoo) island factor denoting island Palmer Archipelago, Antarctica (Biscoe, Dream Torgersen) bill_length_mm number denoting bill length (millimeters) bill_depth_mm number denoting bill depth (millimeters) flipper_length_mm integer denoting flipper length (millimeters) body_mass_g integer denoting body mass (grams) sex factor denoting penguin sex (female, male) year integer denoting study year (2007, 2008, 2009)","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/penguins_orsf.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"Size measurements for adult foraging penguins near Palmer Station, Antarctica — penguins_orsf","text":"Adélie penguins: Palmer Station Antarctica LTER K. Gorman. 2020. Structural size measurements isotopic signatures foraging among adult male female Adélie penguins (Pygoscelis adeliae) nesting along Palmer Archipelago near Palmer Station, 2007-2009 ver 5. Environmental Data Initiative. doi:10.6073/pasta/98b16d7d563f265cb52372c8ca99e60f Gentoo penguins: Palmer Station Antarctica LTER K. Gorman. 2020. Structural size measurements isotopic signatures foraging among adult male female Gentoo penguin (Pygoscelis papua) nesting along Palmer Archipelago near Palmer Station, 2007-2009 ver 5. Environmental Data Initiative. doi:10.6073/pasta/7fca67fb28d56ee2ffa3d9370ebda689 Chinstrap penguins: Palmer Station Antarctica LTER K. Gorman. 2020. Structural size measurements isotopic signatures foraging among adult male female Chinstrap penguin (Pygoscelis antarcticus) nesting along Palmer Archipelago near Palmer Station, 2007-2009 ver 6. Environmental Data Initiative. doi:10.6073/pasta/c14dfcfada8ea13a17536e73eb6fbe9e Originally published : Gorman KB, Williams TD, Fraser WR (2014) Ecological Sexual Dimorphism Environmental Variability within Community Antarctic Penguins (Genus Pygoscelis). PLoS ONE 9(3): e90081. doi:10.1371/journal.pone.0090081","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":null,"dir":"Reference","previous_headings":"","what":"Automatic variable values for dependence — pred_spec_auto","title":"Automatic variable values for dependence — pred_spec_auto","text":"partial dependence individual conditional expectations, function allows variable considered without specify values set variable . values used based quantiles continuous variables (10th, 25th, 50th, 75th, 90th) unique categories categorical variables.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Automatic variable values for dependence — pred_spec_auto","text":"","code":"pred_spec_auto(...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Automatic variable values for dependence — pred_spec_auto","text":"... names variables use. can quotes quotes (see examples).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Automatic variable values for dependence — pred_spec_auto","text":"character vector names","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Automatic variable values for dependence — pred_spec_auto","text":"function used context orsf_pd orsf_ice functions.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/pred_spec_auto.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Automatic variable values for dependence — pred_spec_auto","text":"","code":"fit <- orsf(penguins_orsf, species ~., n_tree = 5) orsf_pd_oob(fit, pred_spec_auto(flipper_length_mm)) #> class flipper_length_mm mean lwr medn upr #> 1: Adelie 185 0.5354109 0 0.50928468 1.0000000 #> 2: Adelie 190 0.5034360 0 0.47752809 1.0000000 #> 3: Adelie 197 0.4458775 0 0.11111111 1.0000000 #> 4: Adelie 213 0.3343601 0 0.07691303 1.0000000 #> 5: Adelie 221 0.3287078 0 0.07691303 1.0000000 #> 6: Chinstrap 185 0.2630916 0 0.03726345 1.0000000 #> 7: Chinstrap 190 0.2596916 0 0.03208729 1.0000000 #> 8: Chinstrap 197 0.2220773 0 0.02031115 1.0000000 #> 9: Chinstrap 213 0.2559835 0 0.03333333 1.0000000 #> 10: Chinstrap 221 0.2269255 0 0.02970297 0.9711538 #> 11: Gentoo 185 0.2014975 0 0.00000000 1.0000000 #> 12: Gentoo 190 0.2368724 0 0.00000000 1.0000000 #> 13: Gentoo 197 0.3320452 0 0.00000000 1.0000000 #> 14: Gentoo 213 0.4096564 0 0.14285714 1.0000000 #> 15: Gentoo 221 0.4443667 0 0.28571429 1.0000000"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":null,"dir":"Reference","previous_headings":"","what":"Prediction for ObliqueForest Objects — predict.ObliqueForest","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Compute predicted values oblique random forest. Predictions may returned aggregate (.e., averaging trees) tree-specific.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"","code":"# S3 method for ObliqueForest predict( object, new_data, pred_type = NULL, pred_horizon = NULL, pred_aggregate = TRUE, pred_simplify = FALSE, na_action = NULL, boundary_checks = TRUE, n_thread = NULL, verbose_progress = NULL, ... )"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"object (ObliqueForest) trained oblique random forest object (see orsf). new_data data.frame, tibble, data.table compute predictions . pred_type (character) type predictions compute. Valid options survival : 'risk' : probability event pred_horizon. 'surv' : 1 - risk. 'chf': cumulative hazard function 'mort': mortality prediction 'time': survival time prediction classification: 'prob': probability class 'class': predicted class regression: 'mean': predicted mean, .e., expected value pred_horizon (double) relevent survival forests. value vector indicating time(s) predictions calibrated . E.g., predicting risk incident heart failure within next 10 years, pred_horizon = 10. pred_horizon can NULL pred_type 'mort', since mortality predictions aggregated event times pred_aggregate (logical) TRUE (default), predictions aggregated trees taking mean. FALSE, returned output contain one row per observation one column tree. length pred_horizon two pred_aggregate FALSE, result list matrices, 'th item list corresponding 'th value pred_horizon. pred_simplify (logical) FALSE (default), predictions always returned numeric matrix list numeric matrices. TRUE, predictions may simplified vector, e.g., pred_type 'mort' survival 'class' classification, array matrices length(pred_horizon) > 1. na_action (character) happen new_data contains missing values (.e., NA values). Valid options : 'fail' : error thrown new_data contains NA values 'pass' : output NA rows new_data 1 NA value predictors used object 'omit' : rows new_data incomplete data dropped 'impute_meanmode' : missing values continuous categorical variables new_data imputed using mean mode, respectively. clarify, mean mode used impute missing values training data object, new_data. boundary_checks (logical) TRUE, pred_horizon checked make sure requested values less maximum observed time object's training data. FALSE, checks skipped. n_thread (integer) number threads use computing predictions. Default 0, allows suitable number threads used based availability. verbose_progress (logical) TRUE, progress messages printed console. FALSE (default), nothing printed. ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"matrix predictions. Column j matrix corresponds value j pred_horizon. Row matrix corresponds row new_data.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"new_data must columns equivalent types data used train object. Also, factors new_data must levels data used train object. pred_horizon values exceed maximum follow-time object's training data, truly want , set boundary_checks = FALSE can use pred_horizon large want. Note predictions beyond maximum follow-time object's training data equal predictions maximum follow-time, aorsf estimate survival beyond maximum observed time. unspecified, pred_horizon may automatically specified value used oobag_pred_horizon object created (see orsf).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"","code":"library(aorsf)"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"classification","dir":"Reference","previous_headings":"","what":"Classification","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Predict probability class predicted class:","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_clsf <- orsf(data = penguins_orsf_train, formula = species ~ .) # predicted probabilities, the default predict(fit_clsf, new_data = penguins_orsf_test[1:5, ], pred_type = 'prob') ## Adelie Chinstrap Gentoo ## [1,] 0.9405286 0.04125900 0.018212368 ## [2,] 0.9628964 0.03459853 0.002505059 ## [3,] 0.9029383 0.08527806 0.011783605 ## [4,] 0.9301983 0.05180907 0.017992625 ## [5,] 0.7968234 0.16538539 0.037791201 # predicted class (as a matrix by default) predict(fit_clsf, new_data = penguins_orsf_test[1:5, ], pred_type = 'class') ## [,1] ## [1,] 1 ## [2,] 1 ## [3,] 1 ## [4,] 1 ## [5,] 1 # predicted class (as a factor if you use simplify) predict(fit_clsf, new_data = penguins_orsf_test[1:5, ], pred_type = 'class', pred_simplify = TRUE) ## [1] Adelie Adelie Adelie Adelie Adelie ## Levels: Adelie Chinstrap Gentoo"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"regression","dir":"Reference","previous_headings":"","what":"Regression","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Predict mean value outcome:","code":"set.seed(329) index_train <- sample(nrow(penguins_orsf), 150) penguins_orsf_train <- penguins_orsf[index_train, ] penguins_orsf_test <- penguins_orsf[-index_train, ] fit_regr <- orsf(data = penguins_orsf_train, formula = bill_length_mm ~ .) predict(fit_regr, new_data = penguins_orsf_test[1:5, ], pred_type = 'mean') ## [,1] ## [1,] 37.74136 ## [2,] 37.42367 ## [3,] 37.04598 ## [4,] 39.89602 ## [5,] 39.14848"},{"path":"https://bcjaeger.github.io/aorsf/reference/predict.ObliqueForest.html","id":"survival","dir":"Reference","previous_headings":"","what":"Survival","title":"Prediction for ObliqueForest Objects — predict.ObliqueForest","text":"Begin fitting oblique survival random forest: Predict risk, survival, cumulative hazard one several times: Predict mortality, defined number events forest’s population observations characteristics like current observation. type prediction require specify prediction horizon","code":"set.seed(329) index_train <- sample(nrow(pbc_orsf), 150) pbc_orsf_train <- pbc_orsf[index_train, ] pbc_orsf_test <- pbc_orsf[-index_train, ] fit_surv <- orsf(data = pbc_orsf_train, formula = Surv(time, status) ~ . - id, oobag_pred_horizon = 365.25 * 5) # predicted risk, the default predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'risk', pred_horizon = c(500, 1000, 1500)) ## [,1] [,2] [,3] ## [1,] 0.013648562 0.058393393 0.11184029 ## [2,] 0.003811413 0.026857586 0.04774151 ## [3,] 0.030548361 0.100600301 0.14847107 ## [4,] 0.040381075 0.169596943 0.27018952 ## [5,] 0.001484698 0.006663576 0.01337655 # predicted survival, i.e., 1 - risk predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'surv', pred_horizon = c(500, 1000, 1500)) ## [,1] [,2] [,3] ## [1,] 0.9863514 0.9416066 0.8881597 ## [2,] 0.9961886 0.9731424 0.9522585 ## [3,] 0.9694516 0.8993997 0.8515289 ## [4,] 0.9596189 0.8304031 0.7298105 ## [5,] 0.9985153 0.9933364 0.9866235 # predicted cumulative hazard function # (expected number of events for person i at time j) predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'chf', pred_horizon = c(500, 1000, 1500)) ## [,1] [,2] [,3] ## [1,] 0.015395388 0.067815817 0.14942956 ## [2,] 0.004022524 0.028740305 0.05424314 ## [3,] 0.034832754 0.127687156 0.20899732 ## [4,] 0.059978334 0.233048809 0.42562310 ## [5,] 0.001651365 0.007173177 0.01393016 predict(fit_surv, new_data = pbc_orsf_test[1:5, ], pred_type = 'mort') ## [,1] ## [1,] 23.405016 ## [2,] 15.362916 ## [3,] 26.180648 ## [4,] 36.515629 ## [5,] 5.856674"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":null,"dir":"Reference","previous_headings":"","what":"Inspect Forest Parameters — print.ObliqueForest","title":"Inspect Forest Parameters — print.ObliqueForest","text":"Printing ORSF model tells : Linear combinations: identified? N observations: Number rows training data N events: Number events training data N trees: Number trees forest N predictors total: Total number columns predictor matrix N predictors per node: Number variables used linear combinations Average leaves per tree: proxy depth trees Min observations leaf: See leaf_min_obs orsf Min events leaf: See leaf_min_events orsf OOB stat value: --bag error fitting trees OOB stat type: --bag error computed? Variable importance: variable importance computed?","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Inspect Forest Parameters — print.ObliqueForest","text":"","code":"# S3 method for ObliqueForest print(x, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Inspect Forest Parameters — print.ObliqueForest","text":"x (ObliqueForest) oblique random survival forest (ORSF; see orsf). ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Inspect Forest Parameters — print.ObliqueForest","text":"x, invisibly.","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.ObliqueForest.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Inspect Forest Parameters — print.ObliqueForest","text":"","code":"object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 5) print(object) #> ---------- Oblique random survival forest #> #> Linear combinations: Accelerated Cox regression #> N observations: 276 #> N events: 111 #> N trees: 5 #> N predictors total: 17 #> N predictors per node: 5 #> Average leaves per tree: 20.8 #> Min observations in leaf: 5 #> Min events in leaf: 1 #> OOB stat value: 0.76 #> OOB stat type: Harrell's C-index #> Variable importance: anova #> #> -----------------------------------------"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":null,"dir":"Reference","previous_headings":"","what":"Print ORSF summary — print.orsf_summary_uni","title":"Print ORSF summary — print.orsf_summary_uni","text":"Print ORSF summary","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Print ORSF summary — print.orsf_summary_uni","text":"","code":"# S3 method for orsf_summary_uni print(x, n_variables = NULL, ...)"},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Print ORSF summary — print.orsf_summary_uni","text":"x object class 'orsf_summary' n_variables number variables print ... arguments passed methods (currently used).","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Print ORSF summary — print.orsf_summary_uni","text":"invisibly, x","code":""},{"path":"https://bcjaeger.github.io/aorsf/reference/print.orsf_summary_uni.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Print ORSF summary — print.orsf_summary_uni","text":"","code":"object <- orsf(pbc_orsf, Surv(time, status) ~ . - id, n_tree = 25) smry <- orsf_summarize_uni(object, n_variables = 2) print(smry) #> #> -- ascites (VI Rank: 1) ------------------------ #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0 0.3087374 0.1859818 0.04183841 0.5614236 #> 1 0.4965395 0.4214549 0.30005793 0.7145065 #> #> -- bili (VI Rank: 2) --------------------------- #> #> |---------------- Risk ----------------| #> Value Mean Median 25th % 75th % #> 0.60 0.2404645 0.1342975 0.03456875 0.3904944 #> 0.80 0.2425520 0.1435297 0.03507037 0.3904944 #> 1.40 0.2625349 0.1554767 0.04820122 0.4168304 #> 3.52 0.3795922 0.3167143 0.15794919 0.5811623 #> 7.25 0.4682454 0.4348054 0.25161269 0.6726243 #> #> Predicted risk at time t = 1788 for top 2 predictors"},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-012-unreleased","dir":"Changelog","previous_headings":"","what":"aorsf 0.1.2 (unreleased)","title":"aorsf 0.1.2 (unreleased)","text":"CRAN release: 2024-01-15 Allowed option \"time\" pred_type predict partial dependence predict survival time (see https://github.com/ropensci/aorsf/issues/37). Added pred_spec_auto() convenient specification variables partial dependence. Partial dependence now runs much faster multiple threads. Added orsf_vint() compute variable interaction scores using partial dependence. Added orsf_update(), can copy modify obliqueForest modify place. Added orsf_control functions classification, regression, survival (https://github.com/ropensci/aorsf/pull/25). optimization implemented matrix multiplication prediction (https://github.com/ropensci/aorsf/pull/20)","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-011","dir":"Changelog","previous_headings":"","what":"aorsf 0.1.1","title":"aorsf 0.1.1","text":"CRAN release: 2023-10-26 Fixed uninitialized value pd_type Fixed various issues related memory leaks","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-010","dir":"Changelog","previous_headings":"","what":"aorsf 0.1.0","title":"aorsf 0.1.0","text":"CRAN release: 2023-10-13 Re-worked internal C++ routines following design ranger. Re-worked progress printed console verbose_progress TRUE, following design ranger. Messages now indicate action taken, % complete, approximate time finishing action. Improved variable importance, following design ranger. Importance now computed tree--tree instead aggregate. Additionally, mortality type prediction used importance survival trees, since mortality depend pred_horizon. Allowed multi-threading performed orsf(), predict.orsf_fit(), functions orsf_vi() orsf_pd() family. Allowed sampling without replacement sampling specific fraction observations orsf() Included Harrell’s C-statistic option assessing goodness splits growing trees. Fixed issue uninformative error message occur pred_horizon > max(time) orsf_summarize_uni. Thanks @JyHao1 @DustinMLong finding !","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-007","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.7","title":"aorsf 0.0.7","text":"CRAN release: 2023-01-12 Additional changes internal testing avoid problems ATLAS","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-006","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.6","title":"aorsf 0.0.6","text":"CRAN release: 2023-01-06 Minor fix internal tests failing run ATLAS","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-005","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.5","title":"aorsf 0.0.5","text":"CRAN release: 2022-12-14 orsf() longer throws errors warnings try give single predictor. note added documentation details ?orsf explains using single predictor orsf() somewhat useless. done resolve https://github.com/mlr-org/mlr3extralearners/issues/259. predict.orsf_fit now accepts pred_horizon = 0 returns sensible values. Thanks @mattwarkentin feature request. added function perform variable selection, orsf_vs(). Made variable importance consistent respect group_factors. Originally, output orsf ungrouped VI values orsf_vi grouped values. update, orsf defaults grouped values. ungrouped values can still recovered. Fixed issue orsf_pd functions output data returned original scale.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-004","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.4","title":"aorsf 0.0.4","text":"CRAN release: 2022-11-07 orsf formulas now accepts Surv objects (see https://github.com/ropensci/aorsf/issues/11) Added verbose_progress input orsf, prints messages console indicating progress. Allowance missing values orsf. Mean mode imputation performed observations missing data. values can also used impute new data missing values. Centering scaling predictors now done prior growing forest.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-003","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.3","title":"aorsf 0.0.3","text":"CRAN release: 2022-10-09 Included rOpenSci reviewers Christopher Jackson, Marvin N Wright, Lukas Burk DESCRIPTION reviewers. Thank ! Added clarification docs pros/cons different variable importance techniques Added regression tests aorsf versus obliqueRSF (similar) Additional support tests functions long right hand sides Updated --bag vignette appropriate custom functions. Allow status values input data general, .e., just 0 1. Allow missing values predict functions, including partial dependence.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-002","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.2","title":"aorsf 0.0.2","text":"CRAN release: 2022-09-05 Modified unit tests compatibility extra checks run CRAN.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-001","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.1","title":"aorsf 0.0.1","text":"CRAN release: 2022-08-23 Added orsf_control_custom(), allows users submit custom functions identifying linear combinations inputs growing oblique decision trees. Added weights input orsf, allowing users fit orsf specific data training set. Added chf mort options predict.orsf_fit(). Mortality predictions fully implemented yet - supported partial dependence --bag error estimates. features added future update.","code":""},{"path":"https://bcjaeger.github.io/aorsf/news/index.html","id":"aorsf-0009000","dir":"Changelog","previous_headings":"","what":"aorsf 0.0.0.9000","title":"aorsf 0.0.0.9000","text":"Core features implemented: fit, interpret, predict using oblique random survival forests. Vignettes + Readme covering usage core features. Website hosted GitHub pages, managed pkgdown.","code":""}]