v0.15.1

This is a bugfix release which contains a number of small improvements to the 0.15.0 release.

What's Changed

• Fix needs_fit logic for model selection with a fixed model by @kbattocchi in #855
• CATE validation - uplift uniform confidence bands by @amarvenu in #840
• Typo fixes by @alexmalins in #864
• Fix shap_values compatibility with shap>=0.43.0 by adjusting check_additivity parameter handling for TreeExplainer by @jcreinhold in #872
• Optimizing NormalInferenceResults confidence interval method speed by @gdaiha in #879
• Enable scikit-learn 1.5 compatibility by @kbattocchi in #891
• A number of improvements to our infrastructure (#861, #863, #867, #876, #880, #882)

New Contributors

• @alexmalins made their first contribution in #864
• @jcreinhold made their first contribution in #872
• @gdaiha made their first contribution in #879

Full Changelog: v0.15.0...v0.15.1

v0.15.0

This is a major update of the library, with several important improvements, as well as a number of bug-fixes and minor improvements.

What's Changed

• Adds the ability to scale ortholearner training using Ray (by @v-shaal in #800)
• Allows missing values in W and sometimes X (by @fverac in #791)
• Enables federated estimation of linear models (by @kgao in #818)
• Provides a CATE validation suite for discrete models (by @amarvenu in #777 and #836)
• Enable model selection for first stage models (by @kbattocchi in #808 and #848)
• Improve discreteness handling, allow binary outcomes by @fverac in #816

New Contributors

• @star1327p made their first contribution in #713
• @v-shaal made their first contribution in #800
• @kgao made their first contribution in #818
• @amarvenu made their first contribution in #777

Full Changelog: v0.14.1...v0.15.0

v0.15.0b1

What's Changed

This beta has many bug fixes and provides several big improvements:

• Adds the ability to scale ortholearner training using Ray (by @v-shaal in #800)
• Allows missing values in W and sometimes X (by @fverac in #791)
• Enables federated estimation of linear models (by @kgao in #818)
• Provides a CATE validation suite for discrete models (by @amarvenu in #777)
• Enable model selection for first stage models (by @kbattocchi in #808)

New Contributors

• @v-shaal made their first contribution in #800
• @kgao made their first contribution in #818

Full Changelog: v0.14.1...v0.15.0b1

v0.14.1

What's Changed

Details

Minor updates and bugfixes

• Avoid returning defaultdict directly when calling est.shap_values @kbattocchi in #709
• Refactor DynamicDML to remove incompatible method signatures by @kbattocchi in #717
• Fix bug when using .score in _OrthoLearner subclasses by @kbattocchi in #760

Package changes

• Enable sklearn 1.2 by @00helloworld and @kbattocchi in #712
• Enable compatibility with pandas 2.0 by @kbattocchi in #750
• Fix numpy 1.24 compatibility by @kbattocchi in #750
• Bump supported shap version limit by @kbattocchi in #750

Build system changes

• Enable GitHub Actions by @kbattocchi in #746
• Use verbose pip install when debugging workflows by @kbattocchi in #746
• Update developer documentation by @kbattocchi in #749

Misc

• Update copyrights and references to point to PyWhy by @kbattocchi in #749
• Updated CI badge in README by @kbattocchi and @fverac #771

Full Changelog: v0.14.0...v0.14.1

v0.14.0

What's Changed

This release contains a major new feature, treatment featurization (#615), plus a number of bugfixes and minor improvements.

Breaking changes: several deprecated features have now been removed, and DynamicDML has been moved to a new econml.panel package.

Details

Major changes

• treatment featurization by @fverac in #615
• Create panel subpackage by @kbattocchi in #662
• Remove deprecated functionality by @kbattocchi in #663

Minor updates and bugfixes

• fix small typo by @Padarn in #651
• initial commit for aka.ms/econml doc migration by @fverac in #640
• Update README.md by @coreynoone in #658
• add usage examples in docs for backdoor and IV methods, add KernelDML to docs by @fverac in #677
• Update Policy Learning with Trees and Forests.ipynb by @OliverTseng in #685

Package changes

• Remove dowhy as core dependency by @kbattocchi in #661
• add matplotlib version ceiling by @fverac in #680

Build system changes

• update some pipeline images (macos-12, ubuntu20), fix pycodestyle E275 by @fverac in #672

New Contributors

• @Padarn made their first contribution in #651
• @coreynoone made their first contribution in #658
• @OliverTseng made their first contribution in #685

Full Changelog: v0.13.1...v0.14.0

v0.13.1

What's Changed

This is a minor release which mainly fixes several minor bugs and compatibility issues with certain versions of other libraries such as dowhy and sklearn.

Details:

Bug fixes

• improve input processing for StatsModelsLinearRegression by @fverac in #599
• Fix RScorer circular import by @fverac in #596
• Docs - typos in dr.rst by @stanton119 in #603
• Fix small issues by @kbattocchi in #617
• Fix typo in motivation docs by @imatiach-msft in #624
• Another minor typo fix in docs by @imatiach-msft in #625
• Replace Boston housing analysis with Ames housing by @fverac in #598
• Several small fixes by @kbattocchi in #629
• Handle pandas categorical types for categorical columns in _causal_analysis.py by @gaugup in #602
• fix a typo by @zjyfdu in #636
• fix _preprocess_data to align with sklearn refactor by @fverac in #638
• Fix RScorer circular import by @fverac in #596
• Docs - typos in dr.rst by @stanton119 in #603

Compatibility changes

• updates dependency dowhy to v0.7 by @xrowan in #568
• upgrade econml to latest version of shap by @imatiach-msft in #595
• Fix support for sklearn<1.0 by @kbattocchi in #637

Build system improvements

• Profiling Unit Tests by @xrowan in #575
• Disable serial tests on macOS by @kbattocchi in #604
• Adding Microsoft SECURITY.MD by @microsoft-github-policy-service in #622

Full Changelog: v0.13.0...v0.13.1

v0.13.0

This release enables support for Python 3.9 and sklearn 1.0 and improves the documentation of the OrthoIV and DRIV classes.

Note that for the moment, the Python 3.9 version of econml does not support the deepiv module because of conflicts with our required versions of tensorflow and keras; we hope to address this in a subsequent release.

What's Changed

• Add notebook for DRIV by @heimengqi in #518

Full Changelog: v0.12.0...v0.13.0

v0.12.0

This release contains several major new features:

• Greatly improved functionality for estimating treatment effects with instrumental variables (#460)
• A variant of DoubleML for use on dynamic data (#446)
• A new class for assessing the causal contributions of different features to an outcome, conditional on the other features (this is spread out over a number of PRs)

There have been a few breaking changes:

• Alpha values for confidence intervals now always default to 0.05; previously some alphas defaulted to 0.1 instead (#482)
• The deprecated ability to pass X, W, and Z to some fit and score methods by position has been removed (#482)

There have also been many bug fixes in this release; we'd particularly like to highlight:

• Default values of alpha are now passed correctly through results classes (#476, thanks @mtanghu!)
• Prevent NaN outputs from GRF in certain corner cases (#480, thanks @arose13!)

v0.12.0b6

This is a beta preparing for our next major release, but does not contain any new user-facing features.

v0.12.0b5

This is a beta preparing for our next major release, but does not contain any new user-facing features.