Releases: giuseppec/iml
Releases · giuseppec/iml
iml 0.11.1
iml 0.11.0
- Fix possible future issue with
data.table::melt()
(#182) - Account for R 4.2.0 changes (#189, @pat-s)
- Add {bit64} to suggests because of {data.table} (#190)
- Test optimizations (#190)
- Enhance documentation for LocalModel (#187, @mirka-henninger)
- Move {keras} from imports to suggests (@pat-s)
- Run tests in parallel (@pat-s)
- Fix off-by-one error in euclidean distance computation (#163, @mirka-henninger)
- Fix
FeatureEffect
handling of empty levels (#160, @grantirv) - Allow computation of importance for groups of features (
FeatureImp
) (#158) FeatureEffect
can now be computed with user provided grid points. Works for ice, ale and pdp.FeatureImp
gets new argumentfeatures
which allows to calculate feature importance for a subset of features. If a list of characters is provided, the joint feature importance per group is calculated (#156, @grantirv)
iml 0.10.1
- Ensure that the
data
argument inPredictor$new()
is always a data.frame (#126) - Fix CRAN checks
iml 0.10.0
- Changes to FeatureEffect$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
v0.8.1
ups version
JOSS submission
adds dois for JOSS paper
v0.5.0
continues paper
v0.4.0
ups version to 0.4.0
v0.2: Merge pull request #6 from christophM/cleanup
Rework API and prepare CRAN release