NEWS

DALEXtra 2.2.1

fixed explain_tidymodels to ignore residual_function for classification models.
fixed explain_h2o examples that might occasionally crash.

bump the requirement for DALEX to 2.4.0.
remove randomForest from suggest due to it enforcing R v4.1 (changed to ranger).
fix predict_surrogate() when new_observation has too many variables (e.g. target outcome).
auto-convert the mlr3 learner-like objects with mlr3::as_learner() in explain_mlr3().
Skip explain_keras and explain_scikitlearn examples while running on macOS as they can rise false-positive errors during R CMD check for some versions of macOS. The very same code still executes properly in tests.
Skip check if the model is trained in explain_tidymodels if the model inherits from model_fit class.
Add support for stacked tidymodels (stacks package).
Add dalex_load_explainer function.
Clear up documentation.

explain_tidymodels() added as a support for tidymodels workflows.
Removed aspect importance. It’s available in triplot package https://cran.r-project.org/web/packages/triplot/index.html.
predict_surrogate() function is added to provide easier interface of accessing lime/iml/localModel implementations of the LIME method.

In added yhat.GraphLearner() and model_info.GraphLearner() to handle GraphLearners mlr3 objects.
New examples.

In explain_h2o() data parameter will bo converted to data.frame if H2OFrame object was passed.
Aspect importance related functions set deprecated. Will be removed with next release.
explain_xgboost() function added

DALEXtra now supports multiclass classification (accordingly to DALEX >= 1.3)
funnel_mesure() and training_test_comparison() recognizes type of the task and applies proper loss_function
yhat.WrappedModel() returns factor response if predict.type is not prob.

Removed h2o::init() from explain_h2o()
Removed mljar support as mljar package is not available for R 3.6.2
Ajusted to DALEX 1.0
fixed yhat.LearnerClassif() returning wrong column of probabilities (PR #34, thanks Hubert!)

Rebuilded plot.overall_comparison() (I lack words that could describe Your greatness, Ania!).
New README and DESCRIPTION. They are more accurate now.
Small fixes to funnel_measure() that imporves it’s stability.

New plot function for funnel_measure() objects. (Thanks Anna Kozak, You are awesome!).
New tests for funnel_measure() and plot.funnel_measure() (Once again You are awesome, Ania!).

explain_keras() added.
explain_mljar() added.
documentation refreshed with links to functions.
explain_scikitlearn() rebuilded. Some of the code was exported to inner functions (helper_functions.R).
conda installation in README.md.
scikitlearn_unix.yml file renamed to testing_environment.yml.

explain_scikitlearn() rebuilded. Now class scikitlearn_model is a additional class for original Python object instead of another object.
explainers created with explain_scikitlearn() have addidtional field param_set.
yhat() is now generic.
New examples in README.md.

Now when you pass .yml that consist environment name that already exists one the machine, DALEXtra will not rise an error and contiune work with existing env.
If condaenv is NULL when creating_env on unixlike OS, DALEXtra will try to find conda on his own.
on_attach() function now checks if conda is installed. Alert is rised if not.

yhat.R created. Predict functions are stored there in order to be more accesible.
explain_h2o() and explain_mlr() rebuilded.

scikitlearn_unix.yml file added to external data. This helps testing using linuxlike OS.
few minor updates in the documentation.
message in create_env() changed.