NEWS.md
predict_parts() functionloss_yardstick() to get_loss_yardstick() and loss_default() to get_loss_default()
loss_one_minus_accuracy() and get_loss_one_minus_accuracy() (#535)plot.model_performance_roc, loss_one_minus_auc and model_performance_auc functions are rewritten to handle repeated predictions (#442)plot function works for list of explanations (if possible) (#424)nieghbour(s) (EN-spelling) were replaced with neighbor(s) (US-spelling) for the consistency and backword compatibility.predict_diagnostics raised error if neighbor value was higer than nrow(explainer$data).predict_function_target_column) to explain function that allows specifying positive class in binary classification tasks (#250).model_diagnostics() returning an error when data is matrix (#355)model_diagnostics() returning an error when y_hat or residuals is of array class (#319)theme_drwhy on Windowsplot.predict_diagnostics now passess ellipsis to plot.ceteris_paribus_explainer
iBreakDown v1.3.1 and ingredients v1.3.1
plot.predict_parts and plot.model_profile (#277).plot.model_profile for multiple profiles (#237).s parameter in yhat.glmnet and yhat.cv.glmnet.model_diagnostics passing wrong arguments to residual_function.hist geometry in plot.model_performance using wrong arugments.model_performance will not work if model_info$type is NULL.N in model_parts (#287).y parameter in explain function.yhat.ranger causing predicts_parts not to plot correctly when task is multiclass.variable_effect is now deprecatedpredict_parts class to objects and plot.predict_parts functionmodel_parts class to objects and plot.model_parts functionpredict_profile function one can specify how grid points shall be calculated, see variable_splits_type (#267).predict_part function has two new options for type: oscillations_uni and oscillations_emp (#267).plot.model_performance function has a new geom="prc" for Precision Recall curve (#273).DALEX now fully supports multiclass classification.explain() will use new residual function (1 - true class probability) if multiclass classification is detected.model_performance() now support measures for multiclass classification.ggpubr from suggests.lossFunction argument is now deprecated in plot.model_performance(). Use the loss_function argument.model_profile color changed to colors_discrete_drwhy(1) which impacts the color of the line in plot.model_profile
loss_name attribute added to loss functions. It will be passed to plot function for objects created with model_parts.model_profile for Accumulated Local rofiles by default use centering (center = TRUE)n_sample argument in model_parts (now it’s N) (#175)variable_profile calls ingredients::ceteris_paribus (#131).variable_response and feature_response moved to variable_effect and now it calls ingredients::partial_dependency (#131).prediction_breakdown moved to variable_attribution and now it calls iBreakDown::break_down (#131).variable_importance, not it calls the ingredients::variable_importance (#131).model_performance (#130).yhat for lrm models from rms packagetheme_drwhy has now left aligned title and subtitle.residuals_distribution calculates now diagnostic plots based on residuals (#143).model_performance calculates several metrics for classification and regression models (#146).plot.model_performance now supports ROC charts, LIFT charts, Cummulative Gain charts, histograms, boxplots and ecdfresiduals_distributon is now individual_diagnostics and produces objects of the class individual_diagnostics_explainers
plot.individual_diagnostics_explainers now plots objects of the class individual_diagnostics_explainers
yhat for caret models now returns matrix instead of data.framemodel_diagnostics new function that plots residuals againes selected variabletitanic_imputed (#113).weights to the explainer. Note that not all explanations know how to handle weights (#118).yhat() and model_info() now support models created with gbm package.colorize in the explain() as requested in (#112).model_info(). It will extract basic irnformation like model package nam version and task type. (#109, #110)update_data() and update_label(). (#114))titanic_imputed as requested in (#104).explain() function now detects if target variable y is present in the data as requested in (#103).pbiecek/DALEX to ModelOriented/DALEX.colors_breakdown_drwhy(), colors_discrete_drwhy() and colors_diverging_drwhy().scikitlearn_model() is removed as it is not working with python 2.7plot.variable_importance_explainer() has now desc_sorting argument. If FALSE then variable importance will be sorted in an increasing order (#41).ingredients and iBreakDown are added to additional features (#72).feature_response() and variable_response() are marked as Deprecated. It is suggested to use ingredients::partial_dependency(), ingredients::accumulated_dependency() instead (#74).variable_importance() is marked as Deprecated. It is suggested to use ingredients::feature_importance() instead (#75).prediction_breakdown() is marked as Deprecated. It is suggested to use iBreakDown::break_down() or iBreakDown::shap() instead (#76).titanic dataset is copied from stablelearner package. Some features are transformed (some NA replaced with 0, more numeric features).DALEX is being prepared for tighter integration with iBreakDown and ingredients.single_variable and single_feature
theme_drwhy().plot.variable_importance_explainer(). Namely bar_width with widths of bars and show_baseline if baseline shall be included in these plots.plot.variable_response_explainer().plot.prediction_breakdown_explainer().apartments_test and HR_test
single_prediction() are now consistent with breakDown::broken(). Specifically, baseline is now 0 by default instead of "Intercept". The user can also specify the baseline and other arguments by passing them to single_prediction (@kevinykuo, #39). WARNING: Change in the default value of baseline.yhat.* functions help to handle additional parameters to different predict() functions.CITATION infoHR and HRTest. Target variable is a factor with three levels. Is used in examples for classification.plot.model_performance() has now show_outliers parameter. Set it to anything >0 and observations with largest residuals will be presented in the plot. (#34)variable_response() to better support of gbm models (c8393120ffb05e2f3c70b0143c4e92dc91f6c823).plot_model_performance() (e5e61d0398459b78ea38ccc980c4040fd853f449).breakDown v 0.1.6.single_variable() / variable_response() function uses predict_function from explainer (#17)explain() function converts tibbles to data.frame when specified as data argument (#15)explain.default() should help when explain() from dplyr is loaded after DALEX (#16)model_performance(), variable_importance(), variable_response(), outlier_detection(), prediction_breakdown(). Old names are now deprecated but still working. (#12)apartments - will be used in examplesvariable_importance() allows work on full dataset if n_sample is negativeplot_model_performance() uses ecdf or boxplots (depending on geom parameter).single_variable() supports factor variables as well (with the use of factorMerger package). Remember to use type='factor' when playing with factors. (#10)explain(). Old version has an argument predict.function, now it’s predict_function. New name is more consistent with other arguments. (#7)xgboost model (#11)variable_dropout() functionexplain() function implementedsingle_prediction() function implementedsingle_variable() function implemented