Skip to content

Latest commit

 

History

History
179 lines (109 loc) · 5.18 KB

CHANGELOG.md

File metadata and controls

179 lines (109 loc) · 5.18 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[0.8.0] - 2024-10-11

Added

  • Added lightning_gp

[0.7.0] - 2024-09-05

Added

  • Added the TDC ADMET benchmarks dataset (with optional dependency pip install molflux[tdc])
  • Turned on pystan tests

[0.6.0] - 2024-08-23

Removed

  • removed support for python 3.8 and 3.9

Changed

  • Enables uv and increases the use of ruff throughout the codebase.
  • Added standard_deviations input argument for all uncertainty metrics except uncertainty_based_rejection
  • Removed the uncertainty_based_rejection metric from the uncertainty suite
  • The typing of featurise_dataset now confirms that it can act on DatasetDict too.
  • Load backend representations from featurisation metadata using stricter unpacking in order to not trigger UserWarnings

Added

  • updated to use uv.

  • Warning if load_from_dict is passed a dictionary with arbitrary keys outside of the expected specification that are ignored when loading a representation.

  • Added map_light features, which are a combination of Morgan, Avalon, Reduced Graph and handcrafted descriptors from rdkit.

  • Added uncertainty support (predict_with_std, predict_with_prediction_interval, and sample) for the ensemble_regressor model.

  • Added an average_features_regressor model that predicts based on the average of the input model features

  • Added GammaConformityScore and ResidualNormalisedScore to mapie_regressor. These should allow for more adaptive prediction intervals

  • Added out_of_sample_r2regression metric

Fixed

  • Lightning logger config sometimes required an explicit config field to be recognised as a logger config; this is no longer the case.

Security

  • Load torch models with weights_only parameter set to True to address potential security concerns

[0.5.0] - 2024-07-11

Added

  • Enable multi-column representations
  • Add linear_split_with_rotation splitting strategy
  • Added a Bayesian ordinal regression model (ordinal_classifier).
  • mapie_regressor now has predict_with_std and sample methods implemented based on a Gaussian approximation for the prediction interval.
  • Added calibration_gap metric
  • Added option for masking inputs by the references
  • v2 featurisation metadata with support for multi-column inputs

Fixed

  • Fixed the dict for matching modules in lightning. Allows many to one matching.
  • model_config is now correctly overridden in LightningModules. Previously a stale config could have been used.
  • Release PyTorch upper bound (previously <2.1).

Changed

  • Compatible with Pydantic v1 & v2
  • Lower pin on botocore / boto3 to help dependency resolution when installed alongside dvc-s3
  • Use class_resolver to simplify and generalise modularity inside Lightning models.
  • model.train will now always accept a validation_data kwarg. If the underlying model implementation doesn't have validation_data in its model._train (or model._train_multi_data), it will be dropped with a warning.
  • Tag format for wrapped models (ensemble_regressor, ensemble_classifier, mapie_regressor, sklearn_pipeline_regressor, sklearn_pipeline_classifier) changed to make clearer which base models are included. The new tag format is of the form '{model.tag}[{base_model.tag}]'.
  • Changed behaviour of Gaussian NLL from summing likelihoods to averaging them

Removed

  • sd parser
  • Deprecate usage of mean_squared_error with root=True

[0.4.0] - 2024-06-03

Removed

  • Drop parameter multi_class and n_jobs for logistic_regressor in anticipation of numpy>=1.7 removal

[0.3.0] - 2024-02-09

Changed

  • Upgraded datasets>=2.17.0 which fixes a problem with flattening indices
  • Removed failure tests for flattening indices

[0.2.0] - 2024-01-31

Changed

  • Updated the spice dataset from 1.1.1 to 1.1.4

Fixed

  • Patch bug with multiproc and Sequence features of fixed length

[0.1.0] - 2023-01-29

Added

  • Added atom_pair from rdkit
  • Added topological_torsion from rdkit
  • Added CovarianceMixin for modelzoo
  • Added separate root_mean_squared_error metric

Changed

  • prediction_internal_coverage from numpy
  • Updated mapie_regressor
  • Strict warnings
  • removed pkg_resources for importlib
  • HF datasets uses trust_remote_code=True by default
  • updated ruff~=0.1.0
  • updated datasets>=2.16.0

Fixed

  • Fixed accuracy metric

Removed

  • Removed pytest-lazy-fixture

[0.0.1] - 2023-12-15

Removed

  • Removed a featuriser

[0.0.0] - 2023-12-15

Added

  • Initial release

[X.Y.Z] - 20YY-MM-DD

(Template)

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.