NEWS (original) (raw)

survex 1.2.0

• added new calculation_method forsurv_shap() called "treeshap" that uses thetreeshap package (#75)
• enable to calculate SurvSHAP(t) explanations based on subsample of the explainer’s data
• changed default kernel width in SurvLIME from sqrt(p * 0.75) to sqrt(p) * 0.75
• fixed error in SurvLIME when non-factorcategorical_variables were provided

survex 1.1.3

• fixed not being able to plot or print SurvLIME results for the cph model sometimes. (#72)
• added global explanations via the SurvSHAP(t) method (seemodel_survshap() function)
• added plots for global SurvSHAP(t) explanations (seeplot.aggregated_surv_shap())
• added Accumulated Local Effects (ALE) explanations (seemodel_profile(..., type = "accumulated"))
• added 2-dimensional PDP and ALE plots (seemodel_profile_2d() function)
• added plot(..., geom="variable") function for plotting PDP and ALE explanations without the time dimension
• new explainers: for flexsurv models and for Python scikit-survival models (can be used with reticulate)
• new plot type for model_survshap() - curves (with functional box plot)
• added diagnostic explanations - residual analysis (seemodel_diagnostics() function)
• added new times generation method "survival_quantiles"and setting it as default (see explain())
• made improvements on the vignettes for the package (seevignette("pdp") andvignette("global-survshap"))
• increased the test coverage of the package
• reduced the number of expensive requireNamespace()calls (#83)

survex 1.0.0

• breaking change: refactored the structure ofmodel_performance_survival object - calculated metrics are now in a $result list.
• added new calculation_method forsurv_shap() called "kernelshap" that usekernelshap package and its implementation of improved Kernel SHAP (set as default) (#45)
• rename old method "kernel" to"exact_kernel"
• added new import (kernelshappackage)
• fixed invalid color palette order in plot feature importance
• fixed predict_parts survshap running out of memory with more than 16 variables (#25)
• added max_vars parameter for predict_parts explanations (#27)
• set max_vars to 7 for every method
• refactored survshap code (#29, #30, #43)
• fixed survshap error when target columns named different than time and status (#44)
• fixed survlime error when all variables are categorical (#46)
• fixed subtitles in feature importance plots (#11)
• added the possibility to set themes withset_theme_survex() (#32)
• added the possibility of plotting multiplepredict_parts() and model_parts() explanations in one plot (#12)
• fixed the x axis of plots (it now starts from 0) (#37)
• added geom_rug() to all time-dependent plots, marking event and censoring times (#35)
• refactored surv_feature_importance.R - change auxiliary columns to include _ in their name. Necessary changes also done to plotting and printing functions. (#28)
• changed default type argument ofmodel_parts() to "difference" (#33)
• refactored integration of metrics (#31)
• changed behaviour of categorical_variables argument inmodel_parts() and predict_parts(). If it contains variable names not present in the variablesargument, they will be added at the end. (#39)
• added ROC AUC calculation and plotting for selected timepoints inmodel_performance() (#22)
• added explanation_label parameter topredict_parts() function that can overwrite explainer label and thus, enable plotting multiple local SurvSHAP(t) explanations. (#47)
• improved the printing of the explainer (#36)
• reduced the default number of time points for evaluation when creating the explainer to 50

survex 0.2.2

• improved and unified API documentation (#2)
• added references to used methods (#5)
• changed the package used to draw complex plots fromgridExtra to patchwork (#7)
• fixed subtitles in plots (#11)
• fixed calculating of ROC curves for classification problems (#17)
• added wrapper function for measures provided bymlr3proba (#10)
• created vignette showing how to use mlr3proba withsurvex
• fixed incompatibility with new ggplot2 version 3.4
• added function for creating integrated versions of time-dependent metrics (#9)
• move ingredients from imports to suggests

survex 0.1.1

• The survex package is now public
• model_parts, model_profile,predict_parts, predict_profile explanations implemented
• C/D AUC, Brier score and (Harrell’s) concordance index performance measures implemented
• Explain methods for survival, ranger,randomForestSRC, censored andmlr3proba packages.