CHANGELOG.md

Changelog

[2.4.0] - 2023-08-16

• Enhancement
  • Remove support for python 3.6 and 3.7.
  • Bumps in joblib, numpy, pandas, scikit-learn, statsmodels, toolz, catboost, lightgbm, shap, xgboost and test auxiliary packages.

[2.3.1] - 2023-04-11

• Bugfix
  • Remove incorrect lightgbm import from common paths

[2.3.0] - 2023-03-28

• Enhacement
  • Bump maximum allowed scikit-learn
  • Move from CircleCI to Github Actions
  • Add optional weight_column argument for evaluators
  • Change default of min_df from 20 to 1 on TfidfVectorizer
  • Include new optional LGBM parameters to lgbm_classification_learner

[2.2.1] - 2022-09-06

• Bug Fix
  • Including a necessary init file to allow the import of the causal cate learners.
  • Fix a docstring issue where the description of causal learners were not showing all parameters.

[2.2.0] - 2022-08-25

• Enhancement
  • Including Classification S-Learner and T-Learner models to the causal cate learning library.
• Bug Fix
  • Fix validator behavior when receiving data containing gaps and a time based split function that could generate empty training and testing folds and then break. The argument drop_empty_folds can be set to True to drop invalid folds from validation and store them in the log.
• Documentation
  • Including Classification S-Learner and T-Learner documentation, also changing validator documentation to reflect changes.

[2.1.0] - 2022-07-25

• Enhancement
  • Add optional parameter return_eval_logs_on_train to the validator function, enabling it to return the evaluation logs for all training folds instead of just the first one
• Bug Fix
  • Fix import in pd_extractors.py for Python 3.10 compatibility
  • Set a minimal version of Python (3.6.2) for Fklearn
• Documentation
  • Fixing some typos, broken links and general improvement on the documentation

[2.0.0] - 2021-12-28

• Possible breaking changes
  • Allow greater versions of catboost, lightgbm, xgboost, shap, swifter (mostly due to deprecation of support to Python 3.5 and older versions). Libraries depending on fklearn can still restrict the versions of the aforementioned libraries, keeping the previous behavior.

[1.24.0] - 2021-12-06

• New
  • Add causal curves summary
• Bug fix
  • Set correct learner name for learners with column_duplicatable decorator

[1.23.0] - 2021-10-29

• New
  • Add common causal evaluation techniques
  • Add methods to debias a dataframe with a treatment T and confounders X

[1.22.2] - 2021-09-01

• Bug fix
  • Remove cloudpickle from requirements

[1.22.1] - 2021-09-01

• Bug fix
  • Remove cloudpickle from parallel_validator

[1.22.0] - 2021-02-09

• Enhancement
  • Add verbose method to validator and parallel_validator
  • Add column_duplicator decorator to value_mapper
• Bug Fix
  • Fix Spatial LC check
  • Fix circleci

[1.21.0] - 2020-10-02

• Enhancement
  • Now transformers can create a new column instead of replace the input
• Bug Fix
  • Make requirements more flexible to cover the latest releases
  • split_evaluator_extractor now supports eval_name parameter
  • Fixed drop_first_column behaviour in onehot categorizer
• New
  • Add learner to calibrate predictions based on a fairness metric
• Documentation
  • Fixed docstrings for reverse_time_learning_curve_splitter and feature_importance_backward_selection

[1.20.0] - 2020-07-13

• Enhancement
  • Now Catboost learner is pickable

[1.19.0] - 2020-06-17

• Enhancement
  • Improve space_time_split_dataset performance

[1.18.0] - 2020-05-08

• Enhancement
  • Allow users to inform a Placeholder value in imputer learner
• New
  • Add Normalized Discount Cumulative Gain evaluator
• Bug Fix
  • Fix some sklearn related warnings
  • Fix get_recovery logic in make_confounded_data method
• Documentation
  • Add target_categorizer documentation

[1.17.0] - 2020-02-28

• Enhancement
  • Allow users to set a gap between training and holdout in time splitters
  • Raise Errors instead of use asserts
• New
  • Support pipelines with duplicated learners
  • Add stratified split method
• Bug Fix
  • Fix space_time_split holdout
  • Fix compatibility with newer shap version

[1.16.1] - 2020-01-02

• Enhancement
  • Increasing isotonic calibration regression by adding upper and lower bounds.

[1.16.0] - 2019-10-07

• Enhancement
  • Improve split evaluator to avoid unexpected errors
• New
  • Now users can install only the set of requirements they need
  • Add Target encoding learner
  • Add PR AUC and rename AUC evaluator to ROC AUC
• Bug Fix
  • Fix bug with space_time_split_dataset fn
• Documentation
  • Update space time split DOCSTRING to match the actual behaviour
  • Add more tutorials(Pydata)

[1.15.1] - 2019-08-16

• Enhancement
  • Now learners that have a model exposes it in the logs as object key

[1.15.0] - 2019-08-12

• Enhancement
  • Make custom_transformer a pure function
  • Remove unused requirements
• New
  • Now features created by one hot enconding can be used in the next steps of pipeline
  • Shap multiclass support
  • Custom model pipeline
• Bug Fix
  • Fix the way one hot encoding handle nans
• Documentation
  • Minor fix flake8 documentation to make it work in other shells
  • Fix fbeta_score_evaluator docstring
  • Fix typo on onehot_categorizer
  • New tutorial from meetup presentation

[1.14.0] - 2019-04-30

• Enhancement
  • Validator accepts predict_oof as argument
• New
  • Add CatBoosting regressor
  • Data corruption(Macacaos)
• Documentation
  • Multiple fixes in the documentation
  • Add Contribution guide

[1.13.4] - 2019-04-25

• Enhancement
  • Add predict_oof as argument to validator

[1.13.3] - 2019-04-24

• Bug Fix
  • Fix warning in placeholder_imputer

[1.13.2] - 2019-04-10

• Bug Fix
  • Fixing missing warner when there is no row with missing values

[1.13.1] - 2019-04-10

• New
  • Add missing warner transformation

[1.13.0] - 2019-04-01

• New
  • Add public version code