- added implementation of aSHAP (aggregated SHAP) and waterfall plot (#519)
- adding a new system for default color schemes (#541)
- added
cross_entropy as model performance measure to multilabel settings #542
- removed the
yardstick dependency
- new vignette added ‘How to use DALEX with the yardstick package?’
- new datasets from World Happiness Report:
happiness_test and happiness_train (#513)
- new datasets from COVID morality:
covid_summer and covid_spring (#513)
- changed URLs in the DESCRIPTION as requested in (#484)
- Fix model_info documentation (#498)
- Support for yardstic metrics (#495)
- Changed default in
explain(colorize=) according to (#473)
- Added explain/yhat support for
partykit (#438)
-
explain() warns if target has more than two values for classification (#418)
- The
plot.model_performance_roc, loss_one_minus_auc and model_performance_auc functions are rewritten to handle repeated predictions (#442)
- The
plot function works for list of explanations (if possible) (#424)
- Order of explainer labels in different plots is the same. To get to this point, orders in
plot.model_performance(..., geom = "histogram" & "boxplot") are reversed (#400)
- Fixed multiclass explainer when data has one column (#405)
- Now explainer handles R functions (#396)
-
predict_parts function handles the N argument natively (#394)
- All encouters of
nieghbour(s) (EN-spelling) were replaced with neighbor(s) (US-spelling) for the consistency and backword compatibility.
- Fixed bug when
predict_diagnostics raised error if neighbor value was higer than nrow(explainer$data).
- Added new parameter (
predict_function_target_column) to explain function that allows specifying positive class in binary classification tasks (#250).
- Fixed
model_diagnostics() returning an error when data is matrix (#355)
- Fixed R package not working with Python Explainer (#318)
- Fixed
model_diagnostics() returning an error when y_hat or residuals is of array class (#319)
- Fixed grid lines in
theme_drwhy on Windows
- Fixed logical values in y rising unnecessery warnings for classification task (#336)
-
plot.predict_diagnostics now passess ellipsis to plot.ceteris_paribus_explainer
- This version requires
iBreakDown v1.3.1 and ingredients v1.3.1
- Fixed
plot.predict_parts and plot.model_profile (#277).
- Fixed
plot.model_profile for multiple profiles (#237).
- External tests for not suggested packages added to gh-actions (#237).
- Extended and refreshed documentation (#237).
- All dontrun statements changed to donttest according to CRAN policy.
- Added value for
s parameter in yhat.glmnet and yhat.cv.glmnet.
- Fixed
model_diagnostics passing wrong arguments to residual_function.
- Fixed aesthetic for
hist geometry in plot.model_performance using wrong arugments.
-
model_performance will not work if model_info$type is NULL.
- Corrected description of
N in model_parts (#287).
- New warning messages for
y parameter in explain function.
- Solved bug in
yhat.ranger causing predicts_parts not to plot correctly when task is multiclass.
-
variable_effect is now deprecated
- fixed typo in
predict_parts_oscillations_emp
- rewrite tests
- added
predict_parts class to objects and plot.predict_parts function
- added
model_parts class to objects and plot.model_parts function
- plot parameters added to the documentation
- Now in the
predict_profile function one can specify how grid points shall be calculated, see variable_splits_type (#267).
- The
predict_part function has two new options for type: oscillations_uni and oscillations_emp (#267).
- The
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.
- Remove
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.
- fixed tests and WARNINGs on CRAN
-
model_profile for Accumulated Local rofiles by default use centering (center = TRUE)
- deprecate
n_sample argument in model_parts (now it’s N) (#175)
-
ingredients and iBreakDown are now imported by DALEX
- updated title for
plot.model_performance (#160).
- in
explain removed check related to duplicated target variable (#164).
-
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).
- updated
variable_importance, not it calls the ingredients::variable_importance (#131).
- updated
model_performance (#130).
- added
yhat for lrm models from rms package
-
theme_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 ecdf
-
residuals_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.frame
-
model_diagnostics new function that plots residuals againes selected variable
- names of functions are changed to be compliant with latest version of the XAI pyramide
- updated
titanic_imputed (#113).
- added
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.
- new dataset
titanic_imputed as requested in (#104).
- the
explain() function now detects if target variable y is present in the data as requested in (#103).
- the DALEX GitHub repository is transfered from
pbiecek/DALEX to ModelOriented/DALEX.
- Examples updated. Now they use only datasets available from DALEX.
- yhat.H2ORegressionModel and yhat.H2OBinomialModel moved to (DALEXtra) and merged into explain_h2o() function.
- yhat.WrappedModelmoved to (DALEXtra) and merged as explain_mlr() function.
- Wrapper for scikit-learn models restored in (DALEXtra) package.
- loss_one_minus_auc function added to loss_functions.R. It uses 1-auc to compute loss. Function created by Alicja Gosiewska.
- Extension for DALEX avaiable at (DALEXtra)
- the
explain() function is more verbose. With verbose = TRUE (default) it prints detailed information about elements of an explainer (#95).
- New support for scikit-learn models via
scikitlearn_model()
- New
yhat functions for mlr, h2o and caret packages (added by Szymon).
-
plot.variable_importance_explainer() has now desc_sorting argument. If FALSE then variable importance will be sorted in an increasing order (#41).
-
pdp, factorMerger and ALEPlot are going to Suggested. (#60). In next releases they will be deprecated.
- added
predict function that calls the predict_function hidden in the explainer object. (#58).
- the
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.
- temporally there is a duplicated
single_variable and single_feature
- Added new
theme_drwhy().
- New arguments in the
plot.variable_importance_explainer(). Namely bar_width with widths of bars and show_baseline if baseline shall be included in these plots.
- New skin in the
plot.variable_response_explainer().
- New skin in the
plot.prediction_breakdown_explainer().
- Test datasets are now named
apartments_test and HR_test
- For binary classification we return just a second column. NOTE: this may cause some unexpected problems with code dependend on defaults for DALEX 0.2.6.
- New versions of
yhat for ranger and svm models.
- Residual distribution plots for model performance are now more legible when multiple models are plotted. The styling of plot and axis titles have also been improved (@kevinykuo).
- The defaults of
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.
- New
yhat.* functions help to handle additional parameters to different predict() functions.
- Updated
CITATION info
- New dataset
HR and HRTest. Target variable is a factor with three levels. Is used in examples for classification.
- The
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)
- Small fixes in
variable_response() to better support of gbm models (c8393120ffb05e2f3c70b0143c4e92dc91f6c823).
- Better title for
plot_model_performance() (e5e61d0398459b78ea38ccc980c4040fd853f449).
- Tested with
breakDown v 0.1.6.
- The
single_variable() / variable_response() function uses predict_function from explainer (#17)
- New names for some functions:
model_performance(), variable_importance(), variable_response(), outlier_detection(), prediction_breakdown(). Old names are now deprecated but still working. (#12)
- A new dataset
apartments - will be used in examples
-
variable_importance() allows work on full dataset if n_sample is negative
-
plot_model_performance() uses ecdf or boxplots (depending on geom parameter).
- Function
single_variable() supports factor variables as well (with the use of factorMerger package). Remember to use type='factor' when playing with factors. (#10)
- Change in the function
explain(). Old version has an argument predict.function, now it’s predict_function. New name is more consistent with other arguments. (#7)
- New vigniette for
xgboost model (#11)
- Support for global model structure explainers with
variable_dropout() function
- DALEX package is now public
-
explain() function implemented
-
single_prediction() function implemented
-
single_variable() function implemented