Changelog

0.14.0 (2025-07-xx)

• Remove polars as a run dependency.
• Make MetaLearner.feature_importances and Explainer.feature_importances return a list of dicts, rather than a list of pandas.DataFrames.

0.13.0 (2025-05-19)

New features

• Add support for polars input in all fit* and predict* functions.

0.12.0 (2025-01-29)

Other changes

• Comply with scikit-learn versions 1.6 and higher.

0.11.0 (2024-09-05)

New features

• Add support for using scipy.sparse.csr_matrix as datastructure for covariates X.

0.10.0 (2024-08-13)

New features

• Add abstract method MetaLearner.predict_conditional_average_outcomes to metalearners.metalearner.MetaLearner.
• Implement RLearner.predict_conditional_average_outcomes for metalearners.rlearner.RLearner.

Bug fixes

• Fix bug in which the metalearners.slearner.SLearner’s inference step would have some leakage in the in-sample scenario.

0.9.0 (2024-08-02)

New features

• Add MetaLearner.init_args.
• Add FixedBinaryPropensity.
• Add MetaLearner._build_onnx to metalearners.MetaLearner abstract class and implement it for TLearner, XLearner, RLearner, and DRLearner.
• Add MetaLearner._necessary_onnx_models.
• Add DRLearner.average_treatment_effect to compute the AIPW point estimate and standard error for average treatment effects (ATE) without requiring a full model fit.

0.8.0 (2024-07-22)

New features

• Add MetaLearner.fit_all_nuisance and MetaLearner.fit_all_treatment.
• Add optional store_raw_results and store_results parameters to MetaLearnerGridSearch.
• Renamed _GSResult to GSResult.
• Added grid_size_ attribute to MetaLearnerGridSearch.
• Implement CrossFitEstimator.score.

Bug fixes

• Fixed a bug in MetaLearner.evaluate where it failed in the case of feature_set being different from None.

0.7.0 (2024-07-12)

New features

• Add optional adaptive_clipping parameter to DRLearner.

Other changes

• Change the index columns order in MetaLearnerGridSearch.results_.
• Raise a custom error if only one class is present in a classification outcome.
• Raise a custom error if there are some treatment variants which have seen classification outcomes that have not appeared for some other treatment variant.

0.6.0 (2024-07-08)

New features

• Implement MetaLearnerGridSearch.
• Add a scoring parameter to MetaLearner.evaluate and implement the abstract method for XLearner and DRLearner.

Other changes

• Increase the lower bound on scikit-learn from 1.3 to 1.4.
• Drop the run dependency on git_root.

0.5.0 (2024-06-18)

• No longer raise an error if feature_set is provided to SLearner.
• Fix a bug where base model dictionaries – e.g., n_folds or feature-set – were improperly initialized if the provided dictionary’s keys were a strict superset of the expected keys.

0.4.2 (2024-06-18)

• Ship license file.

0.4.1 (2024-06-18)

• Fix dependencies for pip.

0.4.0 (2024-06-18)

• Implemented CrossFitEstimator.clone.
• Added n_jobs_base_learners to MetaLearner.fit.
• Renamed Explainer.feature_importances. Note this is a breaking change.
• Renamed MetaLearner.feature_importances. Note this is a breaking change.
• Renamed Explainer.shap_values. Note this is a breaking change.
• Renamed MetaLearner.shap_values. Note this is a breaking change.
• Renamed MetaLearner.explainer. Note this is a breaking change.
• Implemented synchronize_cross_fitting parameter for MetaLearner.fit.
• Implemented cv parameter for CrossFitEstimator.fit.

0.3.0 (2024-06-03)

• Implemented Explainer with support for binary classification and regression outcomes and discrete treatment variants.
• Integration of Explainer with MetaLearner for feature importance and SHAP values calculations.
• Implemented model reuse through the fitted_nuisance_models and fitted_propensity_model parameters of MetaLearner.
• Allow for fit_params in MetaLearner.fit.

0.2.0 (2024-05-28)

Beta release with:

• DRLearner with support for binary classification and regression outcomes and discrete treatment variants.
• Generalization of TLearner, XLearner, and RLearner to allow for more than two discrete treatment variants.
• Unification of shapes returned by predict methods.
• simplify_output and metalearner_factory.

0.1.0 (2024-05-16)

Alpha release with:

• TLearner with support for binary classification and regression outcomes and binary treatment variants.
• SLearner with support for binary classification and regression outcomes and discrete treatment variants.
• XLearner with support for binary classification and regression outcomes and binary treatment variants.
• RLearner with support for binary classification and regression outcomes and binary treatment variants.