2.0.2

fix bug when EnetGLM is fitted as pure ridge, statsmodels returns the regularized wrapper without the refit option. Therefore, the summary and statistics are not available.

2.0.1

Better docstrings, better documentation and change the web doc style

2.0.0

In this release, I've introduced some exciting new features and enhancements:

• Introducing the option to use fasttreeshap implementation to compute shap importance, offering improved performance. Please note that fasttreeshap does not currently support XGBoost.
• We've added a powerful new feature: lasso feature selection, ideal for models without interactions like LM, GLM, and GAM.
• You can now pass lightgbm parameters to GrootCV, providing more flexibility in customizing your models.
• We've fixed a bug related to sample weight shape in mrMR.
• The latest tutorials and notebooks have updated and enhanced the documentation.

These updates will make your experience with our library even more productive and enjoyable. Happy modeling!

1.1.4

Update tqdm to auto-mode and update the python version in the requirements to avoid dependency issues

1.1.2

Use data type dict for consistency

1.1.0

Complete refactoring and fixing remaining bugs when using catboost

What's Changed

• Fixed handling of categorical features. by @Tialo in #20
• Refactor binning method for better readability and maintainability by @ThomasBury in #21
• Fixed catboost and added new tqdm by @Tialo in #22

New Contributors

• @Tialo made their first contribution in #20

Full Changelog: 1.0.7...1.1.0

1.0.7

Prevent zero division in the conditional entropy

1.0.6

Fixing a bug in mrmr (returning self), preventing errors in sklearn pipeline.

1.0.5

patching classes where old unused argument was causing an error

1.0.2

The first version of ARFS:

• Parallelization of functions applied on pandas data frame
• Faster and more modular association measures
• Removing dependencies (e.g. dython)
• Better static and interactive visualization
• Sklearn selectors rather than a big class
• Discretization of continuous and categorical predictors
• Minimal redundancy maximal relevance feature selection added (a subset of all relevant predictors), based on Uber's MRmr flavor
• architecture closer to the scikit-learn one