0.10 (In Progress)

Changes to FeatureEffect$results data.frame:
- All numeric values are stored in .value (no more .y.hat and .ale)
- Additional column .type denoting the Feature Effect type
- renamed column .feature to .borders. This column stores the grid borders which were used for computing the FE with respect to the used method
- For ALE type the column containing the feature name was moved to the front (as it is for the other types)
Adds support for h2o
Adds support for keras
Adds support for mlr3
Fixes problem with FeatureImp that caused unused features to get non-zero importances
FeatureEffects$plot() based on patchwork now
Use new roxygen R6 documentation
Use futures for parallel code execution
Test on GitHub Actions instead of Travis
Add a pkgdown site
Use “callr” backend in “parallel” vignette

iml 0.9.0

Removes the run parameter from all interpretation methods.
Adds class FeatureEffects which wraps FeatureEffect and allows to compute feature effects for all features of a model with one call.
Add column “.type” to $result data.frame of FeatureEffect when method="ale" and the feature is categorical
Adds parameter ylim to FeatureEffect$plot to manually set the limits of the y-axis for feature effect plots with one feature.
Adds predict method to FeatureEffect, which predicts the marginal effect for data instances.

Some bigger changes in the feature importance class FeatureImp:
- The method argument was removed, only shuffling is now possible. This means the cartesian product of all data points with all data points is not an option any longer. It was never really practical to use, except for toy examples.
- The importance plot shows the name of the loss function in the x-axis label.
- The importance plot shows the quantiles of importance over the different repetitions.
- Default number of repetitions increased to 5.
Fixes problems with missing centering of ALE plots when using multiclass
Automatically extracts data and target from the model when possible (based on the prediction::find_data function). Data extraction doesn’t work with mlr, but target extraction does.
Feature importance (FeatureImp) automatically returned the ratio of permuted model error and original model error. With 0.7.2 the user can choose between the ratio (default) and the difference.

Fixes problems with wrong computation of feature importance, features effects and so on for xgboost models.

The Partial class is deprecated and will be removed in future versions. You should use FeatureEffect now. Its usage is similar to Partial but the aggregation and ice argument are now combined in the new method argument, where you can choose between ‘ale’, ‘pdp’, ‘ice’, ‘pdp+ice’.
Introduced ALE plots into the FeatureEffect class (method='ale'). They are now the default instead of PDPs, because they are faster and unbiased.
Plot for categorical features in PDP changed. Now showing bar plots instead of boxplots when method='pdp'

Removed losses: f1, logLoss, rmse, mdae, rae, rmse, rmsle, rse, rrse f1 because the implementation used didn’t make sense anyways
Interaction: The results return as interaction strength now the H-statistic instead of the H-squared-statistic. This makes it more coherent with the gbm pacakge and the interact.gbm function and with what Friedman uses in the plots in the paper. For users of the package this means that an interaction of strength x becomes an interaction of strength sqrt(x).
Interaction, FeatureImp and Partial are now computed batch-wise in the background. This prevents this methods from overloading the memory. For that, the Predictor has a new init argument ‘batch.size’ which limits the number of rows send to the model for prediction for the methods Interaction, FeatureImp and Partial.
Interaction and FeatureImp additionally allow parallel computation on multiple cores. See vignette("parallel", package = "iml") for how to use it.

The Predictor can be initialized with a type (e.g. type = "prob"), which is more convenient than writing a custom predict.fun. For caret classification models, the default is now to return the response, so make sure to initialize the Predictor with type = "prob" for fine-grained results.
It’s easier to use classifier that output class labels and no probabilities. No warning will be issued anymore. Internally, the class labels are treated as probabilities (one column per class), where the probability for the predicted class is 1, for the others 0.
FeatureImp supports the n.repetitions parameter which controls the number of repetitions of the feature shuffling.

Implemented Interaction measure
Removed feature.index variable from Partial and renamed .class.name column in results to .class.

object$run() does not return self any longer. This means using object$set.feature() for example does not automatically print the object summary any longer.
Added an introductory vignette.
Fixed an issue where the Predictor would not store X, when y is given as character.
The column names of the data.frames with the results of the interpretation methods start with “.” instead of “..”. This is due to a recent change in the data.table package v1.10.5 news item 18.
Removed the deprecated classes PartialDependence and Ice. Use Partial instead.

The API has been reworked:
- User directly interacts with R6 classes (pdp() is now PartialDependence$new()).
- User has to wrap the machine learning model with Predictor$new().
- New data points in Shapley and LocalModel can be set with $explain().
- Lime has been renamed to LocalModel.
Plots have been improved.
Documentation has been improved.

Initial release