- Remove dependency pkgmaker
- Remove maximize_spline_metric and minimize_spline_metric
- Add boot_digits parameter to cutpointr:::print.summary_cutpointr
- The (very long) vignette has been split up into multiple vignettes, each with its own topic. @kapsner has written an additional vignette on bootstrapping.
- Update ?multi_cutpointr to say that it uses cutpointr(), not cutpointr_()
- When using an extra parameter for the metric, such as in sens_constrain, and if bootstrapping was run, that parameter was not forwarded to the bootstrapping
- Clarified the help for the boot_stratify parameter
- Better examples for the constrained metrics
- Added reference and citation for the article about cutpointr in JSS
boot_ci
now works with multiple cutpoints (multiple cutpoints are possible ifbreak_ties = c
).add_metric
now adds the selected metrics to the bootstrap results, too.- Include metrics that were added with
add_metric
insummary()
. - Change default value of
subgroup
inmulti_cutpointr
toNULL
(instead of missing) to make it consistent withcutpointr
. - No rounding anymore within the internal function
summary_sd
so that the various summary functions now return all values without rounding. - Descriptive statistics in summary functions are now stored with the additional first column "Data" instead of giving the class values as row names.
- Nicer printing of summary objects in Rmd documents.
boot_stratify
is now passed to the method functions so that the bootstrap withinmaximize_boot_metric
andminimize_boot_metric
can be stratified, too.- Subtitles (such as "by class") have been removed from the plots because the subtitles should have read "by subgroup" and because this is already clear from the legend.
- Fix a bug in
multi_cutpointr
that forced theclass
variable to be named "suicide".
- Reduce size of tarball for CRAN by removing some superfluous files
- Accelerate tests and vignette building to lower runtime on CRAN
- Some changes for vctrs 0.3.0 and dplyr 1.0.0: Unname and correctly assign classes to a few objects instead of using vctrs::df_cast and vctrs::df_ptype2 to keep compatibility with vctrs 0.2.4
- Added ORCID and article reference to Description
- Minor internal changes for compatibility with the latest updates of
some packages from the tidyverse (e.g.
tibble
3.0.0)
- Prepare for matrix inheriting from class "array" in R 4.0.0 by making a minor
change in non-exported utility function
sanitize_metric
cutpointr
androc
now both use tidyeval.!!
can be used when an argument should be unquoted, as indplyr
, e.g.myvar <- "dsi"; cutpointr(suicide, !!myvar, suicide)
.cutpointr_
is now deprecated. Transforming variables directly in the call is thus no longer supported, e.g.cutpointr(suicide, dsi * 2, suicide)
now throws an error.- The object returned by
multi_cutpointr
does not have thecutpointr
class anymore.
- A new
boot_ci
function is available that calculates confidence intervals (the empirical quantiles) based on the bootstrap results. - The
auc
function is now exported and can be used to calculate the AUC from acutpointr
orroc_cutpointr
object, e.g.auc(roc(suicide, dsi, suicide, "yes", "no"))
boot_test
is a new function for carrying out a bootstrap test for equivalence of a metric, e.g. the AUC, the Youden-Index or also the optimal cutpoint. The standard deviation is calculated assd
of the differences in metric values per bootstrap repetition, then a z-test is calculated.
- New
type
argument toplot_roc
for choosing line or step - The resulting object from
roc_cutpointr
can now simply be plotted withplot()
- The bootstrapping no longer tries to redraw bootstrap samples if only one
class is drawn. In that case the repetition is removed from the results via
.errorhandling = "remove"
inforeach
. - Subsequently report the number of missing values in the bootstrap results.
summary.cutpointr
andsummary.multi_cutpointr
now print an additionalNAs
column in the bootstrap summary andcutpointr
issues a message if any bootstrap repeats failed (e.g. because only one class was drawn). - Stratified bootstrapping is now supported via the
boot_stratify
argument.
- Make the printed output of
summary.cutpointr
andsummary.multi_cutpointr
more compact - No rounding of numbers in
summary.cutpointr
andsummary.multi_cutpointr
any more. The rounding is now done inprint.summary_cutpointr
andprint.summary_multi_cutpointr
, respectively, and can be controlled via thedigits
argument plot_metric
has a newadd_unsmoothed
argument for adding the unsmoothed metric values to the plot as a dashed line (defaultTRUE
). Helpful to inspect the smoothing of functions likemaximize_gam_metric
.- Add some mathematical details to
?oc_youden_kernel
. - The Readme and vignette have been updated and condensed a bit.
- A warning is issued if in
metric_constrain
or one of the other constrained metricsmin_constrain
can not be achieved.
- Fix the default for
break_ties
incutpointr.default
by setting it tomedian
as it was already incutpointr.numeric
andcutpointr_
.
- Let
roc()
return a tibble instead of a data.frame - Printing results of
roc()
is now possible withplot_roc()
- Extra metric columns can now be added to a
roc_cutpointr
object withadd_metric()
- Add prostate_nodal data set of nodal involvement and acid phosphatase levels in 53 prostate cancer patients
- Fix fetching of method name if method was called using
::
or:::
- Make test of summary printing more tolerant after problems with
tidyr
0.8.3 - Issue an error if plot is used on a
multi_cutpointr
object - Add a summary method for
multi_cutpointr
, a correspondingsummary_multi_cutpointr
class and a printing method for that class - The column
variable
is not returned anymore bymulti_cutpointr
, because it is identical topredictor
- Run
multi_cutpointr
only on all numeric columns, ifx = NULL
- Add constrainable metrics, e.g. sens_constrain to calculate sensitivity given a minimum value for specificity
- Fix a bug where dot-arguments were not passed to the metric function in cutpointr_internal
- Add a check to ensure that the metric function does not return only missing values
- Replace (fix) ">" by ">=" in the documentation of
cutpointr()
.
- Add
sigfig
argument toprint.cutpointr
to allow for specifying the number of significant digits to be printed - Add
add_metric()
function to add further metrics to the output ofcutpointr()
- Add
roc01
metric function to calculate the distance of points on the ROC curve to the point (0,1) on ROC space - Fix
plot_sensitivity_specificity()
ifboot_runs = 0
- Fix display of bootstrap results in summary
- Update benchmarks in Readme
- More tests and modified two tests that led to errors when CRAN checked
the package (
spar = NULL
inmaximize_spline_metric
) - Add links to the documentation of all method, plotting, main and metric functions to all other functions of the same family
- Add some checks to predict.cutpointr to prevent improper
cutpoint_nr
- The
boot
column is now always returned andNA
, if no bootstrapping was run, so that the number of returned columns is constant
- Fix a bug in check_method_cols that occurred when a user-supplied method function returned a metric column but no roc_curve
- Add chapters to the Readme about GAM, kernel and normal methods
- Fix bug in bootstrapping that allowed resamples with only one class to be sampled
use_midpoints
is now also passed tomethod
bycutpointr
to allow for the calculation of midpoints withinmaximize_boot_metric
andminimize_boot_metric
, which before happened incutpointr
, leading to slightly biased cutpoints in certain scenarios
- Change defaults in spline smoothing:
nknots
is now calculated bystats::.nknots_smspl
andspar = 1
- Add
cutpoint_tol
argument to define a tolerance around the optimized metric, so that multiple cutpoints in the vicinity of the target metric can be returned and to avoid not returning other "optimal" cutpoints due to floating-point problems - Change default value to
break_ties = c
- If multiple optimal cutpoints are found and they are summarized using
break_ties
, the returned main metric is now not the optimal one but the one corresponding to the summarized cutpoint (thus may be worse than the optimal one) - Add
maximize_gam_metric
andminimize_gam_metric
for smoothing via generalized additive models - All plots with
geom_ribbon
now usesize = 0
to plot no lines around the (transparent) areas
- Return multiple optimal cutpoints
- Add spline smoothing method
- Reformat docs
- Remove superfluous check in
plot_cutpointr
- Add metrics
plr
(positive likelihood ratio),nlr
(negative likelihood ratio),false_discovery_rate
, andfalse_omission_rate
- In preparation for a new version of tibble that will restore the old printing behaviour and in order to pass the CRAN check the print method for cutpointr has been altered depending on the loaded version of tibble.
silent
argument for roc().- Reformat docs to include better formatted equations
- Remove superfluous check in plot_cutpointr
cutpointr_
now accepts functions instead of character strings asmethod
ormetric
- Fix naming the metric in the bootstrap if the supplied function returns an unnamed vector
- Fix help for the
use_midpoints
parameter. If TRUE (default FALSE) the returned optimal cutpoint will be the mean of the optimal cutpoint and the next lowest observation (fordirection = ">="
) or the next highest observation (fordirection = "<="
) - Rename
sum_ppvnpv
,prod_ppvnpv
, andabs_d_ppvnpv
tosum_ppv_npv
,prod_ppv_npv
, andabs_d_ppvnpv
to match the naming scheme to the names of the metrics that optimize sensitivity and specificity - Also all metric names that are returned by the metric functions are now lower case
- The
summary_sd
function now also returns 5% and 95% percentiles that are included in the output ofsummary
- The default number of bootstrap repeats in
minimize_boot_metric
andmaximize_boot_metric
was changed from 200 to 50 - The message "Running bootstrap..." is now displayed before executing the bootstrap so that possible messages or warnings can be attributed to the main cutpoint estimation or cutpoint estimation during the bootstrapping
- The
summary
function now returns a data.frame instead of a list, also the printing method forsummary_cutpointr
has been slightly modified - Add
plot_sensitivity_specificity
for plotting cutpoints vs. sensitivity and specificity on the y-axis - Remove
oc_optimalCutpoints
function - Remove dependencies
ROCR
andOptimalCutpoints
by rewriting tests and storing benchmark results
- cutpointr() now also works on vectors of raw data, that is without the
data
argument. Thus, it can be used as before by specifyingdata
,x
, andclass
or alternatively without specifyingdata
and directly supplying the vectors of predictions and outcomes asx
andclass
. silent
argument for optionally suppressing messages (e.g. which class is assumed to be the positive one)- Parts of the code were rewritten using Rcpp which leads to roughly a threefold speedup with larger n (n > 1e5)
- The automatic determination of the positive / negative classes and direction now uses the median of the predictor values per class instead of the mean so that the AUC is always >= 0.5
- The bootstrapping now also returns the in-bag main metric, the out-of-bag ROC curve, and the out-of-bag AUC
- Add multi_cutpointr for running cutpointr_ on multiple predictor variables
- Add metric risk_ratio
- Use geom_bar instead of geom_histogram if all cutpoints are integer in plot_x and plot_cut_boot or if there is just one unique value
- If bootstrapping is run, a bootstrapped confidence interval can be displayed in plot_metric
- Add methods for median and mean as cutpoint
- Fix display of cutpoint in ROC curve (use get_opt_ind in plot.cutpoint and plot_roc)
- Fix picking correct metric in plot_metric_boot