Changelog

This changelog follows the great advice from https://keepachangelog.com/.

Each section will have a title of the format X.Y.Z (YYYY-MM-DD) giving the version of the package and the date of release of that version. Unreleased changes i.e. those that have been merged into master (e.g. with a .dev suffix) but which are not yet in a new release (on PyPI) are added to the changelog but with the title X.Y.Z (unreleased). Unreleased sections can be combined when they are released and the date of release added to the title.

Subsections for each version can be one of the following;

  • Added for new features.

  • Changed for changes in existing functionality.

  • Deprecated for soon-to-be removed features.

  • Removed for now removed features.

  • Fixed for any bug fixes.

  • Security in case of vulnerabilities.

Each individual change should have a link to the pull request after the description of the change.

0.3.1 (2021-10-05)

Added

  • Add support for absolute error conformal predictions with lgb.Booster models with the LGBMBoosterAbsoluteErrorConformalPredictor class #23

  • Add bandit into build and test github action #22

  • Add bandit to test dependencies in pyproject.toml #22

  • Add badges to README #21

  • Add new github action to check that _version.py and CHANGELOG.rst files are modified in pull requests to the master branch. This workflow is a slightly modified version of mwcodebase/versioning-checker (source code ) #20

Changed

  • Rename .github/workflows/python-package.yml to .github/workflows/build-test.yml #22

  • Change build-test github action to also run on pushes to master #22

  • Change Python package build and test workflow to only trigger on pull requests to master #20

  • Change logo to; logo #14

0.3.0 (2021-09-19)

Added

  • Add logos; old_logo, old_logo_no_tree to README and docs #12

  • Add changelog into sphinx docs #11

  • Add new ConformalPredictor abstract base class that all other conformal predictor classes will inherit from #9

    • Add _lookup_baseline_interval` method in ConformalPredictor which returns the baseline_interval attribute but which can be overridden by the split conformal predictor classes or future classes where the baseline interval is not a constant value

  • Add new tests; TestConformalPredictionValues for the model type specific ConformalPredictor subclasses that test (when using a non-trivial model) #9

    • The conformal predictor is calibrated at the expected level for different values of alpha

    • The conformal predictor gives the expected intervals

  • Add new docstrings.combine_split_mixin_docs function to combine docstring for SplitConformalPredictorMixin and the model specific classes it will be jointly inherited with #9

Changed

  • Update changelog to follow structure recommendations from https://keepachangelog.com/ #11

  • Change file type of chaneglog to .rst #11

  • Update AbsoluteErrorConformalPredictor and LeafNodeScaledConformalPredictor classes to inherit from ConformalPredictor #9

  • Remove predict_with_interval and _calibrate_interval` methods from LeafNodeScaledConformalPredictor class, these are now in the ConformalPredictor class #9

  • Refactor SplitConformalPredictor into SplitConformalPredictorMixin that does not inherit from LeafNodeScaledConformalPredictor #9

    • Rename baseline_intervals attribute to baseline_interval

    • Remove predict_with_interval method

    • Remove calibrate method

  • Revert the nonconformity.nonconformity_at_alpha function to use np.quantile but with interpolation="higher" to select the upper value if the quantile falls between two values #9

  • Abstract out calculation of nonconformity scores into a _calculate_nonconformity_scores` method which is implemented in AbsoluteErrorConformalPredictor and LeafNodeScaledConformalPredictor classes #9

  • Change _sum_dict_values to be a staticmethod of LeafNodeScaledConformalPredictor rather than a function in pitci.base #9

  • Change linting, tests and mypy to always run in the github actions pipeline #9

  • Renamed LGBMBoosterLeafNodeSplitConformalPredictor to LGBMBoosterSplitLeafNodeScaledConformalPredictor #9

  • Renamed XGBoosterLeafNodeSplitConformalPredictor to XGBoosterSplitLeafNodeScaledConformalPredictor #9

  • Renamed get_leaf_node_split_conformal_predictor to get_split_leaf_node_scaled_conformal_predictor #9

0.2.0 (2021-07-26)

Added

  • Add train_data argument to the calibrate methods of XGBoosterLeafNodeScaledConformalPredictor and XGBSklearnLeafNodeScaledConformalPredictor classes to allow the user to calibrate the leaf node counts on a different (train) data sample, rather than the sample used to calibrate the interval widths (which shouldn’t be the training sample) #3

  • Add LGBMBoosterLeafNodeScaledConformalPredictor class to provide leaf node count scaled conformal intervals for lgb.Booster models #4

  • Add sphinx documentation for package in docs folder #5

  • Add SplitConformalPredictor class that allows conformal intervals to be calibrated for different bands of the data based off the scaling factor #6

  • Add XGBoosterLeafNodeSplitConformalPredictor class that allows split conformal intervals with xgb.Booster objects where the scaling factor is based off the leaf node counts #6

  • Add LGBMBoosterLeafNodeSplitConformalPredictor class that allows split conformal intervals with lgb.Booster objects where the scaling factor is based off the leaf node counts #6

  • Consolidate docstrings across inherited classes with new docstrings module #7

Changed

  • Remove xgboost and add pandas to requirements.txt #4

  • Swap project to use flit as the package build tool #7

  • Change calculation of alpha at given quantile to select closest observation if the quantile falls between two values #7

0.1.1 (2021-05-06)

Added

  • Add support for xgb.XGBRegressor and xgb.XGBClassifier objects with non scaled nonconformity measure in XGBSklearnAbsoluteErrorConformalPredictor class #1

  • Add support for xgb.XGBRegressor and xgb.XGBClassifier objects with leaf node scaled nonconformity measure in XGBSklearnLeafNodeScaledConformalPredictor class #1

  • Add dispatches module with helper functions get_absolute_error_conformal_predictor and get_leaf_node_scaled_conformal_predictor to return correct conformal predictor class given the type of underlying model passed #1

Changed

  • Change AbsoluteErrorConformalPredictor to be abstract base class #1

  • Add XGBoosterAbsoluteErrorConformalPredictor class to provide non scaled conformal intervals for xgb.Booster objects, previously the AbsoluteErrorConformalPredictor class provided this functionality #1

  • Change LeafNodeScaledConformalPredictor to be abstract base class #1

  • Add XGBoosterLeafNodeScaledConformalPredictor class to provide leaf node scaled conformal intervals for xgb.Booster objects, previously the LeafNodeScaledConformalPredictor class provided this functionality #1

0.1.0 (2021-05-01)

Added

  • Add AbsoluteErrorConformalPredictor class implementing non scaled conformal intervals for xgb.Booster objects

  • Add LeafNodeScaledConformalPredictor class implementing conformal intervals scaled by leaf node counts for xgb.Booster objects