Old Oaks

Add import methods package for use outside of R GUI/R Studio.

Weeping Willows

This release cleans up the findVariation and variationAcrossGroups functions. findVariation's default behavior is unchanged; variationAcrossGroups produces a slightly different plot by default and eliminates the need for an interactive session.

Elastic Elms

Add "Limone", a lime-like tool for local interpretation of black-box models.

Citrus Blast

Added:

• Kmeans clustering
• XGBoost multiclass support
• findingVariation family of functions

Changed:

• Develop step trains and saves models
• Deploy no longer trains. Loads and predicts on all rows.
• SQL uses a DBI back end

Removed:

• testWindowCol is no longer a param.
• SQL reading/writing is outside model deployment.

Splendid Lemons

Added

• Added getters for predictions getPredictions() in development (lasso, random forest, linear mixed model)
• Added getOutDf to each algorithm deploy file so predictions can go to CSV
• Added percentDataAvailableInDateRange, to eventually replace countPercentEmpty
• Added featureAvailabilityProfiler

Changed

• TimeStamp column predictive output is now local time (not GMT)

Fixed

• Lasso deployment was defaulting to always training a new model

v0.1.11

• Preparation for CRAN including Travis-CI for Linux and OSX
• Bug fixes and example changes from PR #211
• No API changes.

v0.1.10

This is the first full release of healthcare.ai for R. Note that

This release encompasses basic healthcare ML functionality:

• Model comparison between random forest, lasso, and mixed model algorithms
• Feature selection via lasso and random forest feature importance
• Model deployment to SQL Server, providing top-three most important features
• Imputation (column mean for numeric and column mode for categorical)
• Hyperparameter tuning using mtry and number of trees for random forest
• ROC and PR Curves plotted
• Model performance evaluated via AU_ROC and AU_PR
• To assist, these functions are available:
  • groupedLOCF (for longitudinal imputation)
  • findTrends (for Nelson rule 3)
  • convertDateTimeColToDummies to create data-based features from datetime stamp
  • calculateAllCorrelations for correlations across all numeric cols in data frame
  • calculateTargetedCorrelations for correlations across numeric cols and specific column

For this release the following infrastructure: is in place:

AppveyorCI
Roxygen2
mkdocs for website (which files reside in documentation repo)