Skip to content

CHANGELOG.md

Latest commit

 

History

History
152 lines (99 loc) · 5.16 KB

CHANGELOG.md

File metadata and controls

152 lines (99 loc) · 5.16 KB
Raw

Changelog

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

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

[0.1.0-alpha.21] - 2026-03-14

Added

  • Automated CI workflow (.github/workflows/sklearn-compat.yml) to test against scikit-learn 1.6.1, 1.7.2, and 1.8.0 on every push to main, weekly, and on demand
  • README compatibility matrix listing tested scikit-learn versions with a call for users to report incompatibilities

Fixed

  • AttributeError when serializing SimpleImputer on scikit-learn < 1.8.0: _fill_dtype (introduced in 1.8.0) is now skipped gracefully via a hasattr guard, preserving compatibility across all supported versions

[0.1.0-alpha.20] - 2025-10-01

Added

  • High-level save() and load() methods on SerializationManager for convenient file I/O
  • README example for custom estimator support

Changed

  • Minor internal refactoring: removed redundant code and enforced UTF-8 encoding for text mode I/O

[0.1.0-alpha.19] - 2025-06-01

Added

  • Support for custom and third-party estimators via custom_estimators parameter on SklearnSerializer
  • README example showing integration with chemotools pipelines

[0.1.0-alpha.16] - 2025-01-01

Added

  • Taskfile for standardised developer workflows (test, lint, format, type-check, build, etc.)
  • Python 3.13 added to CI matrix
  • Code coverage reporting via codecov

Changed

  • Moved SklearnSerializer to its own subfolder (openmodels/serializers/sklearn/) for better organisation
  • Stopped tracking poetry.lock in version control

[0.1.0-alpha.14] - 2024-11-01

Added

  • Extended scikit-learn estimator support:
    • TargetEncoder, SplineTransformer (scipy BSpline), IsolationForest
    • NeighborhoodComponentsAnalysis, LatentDirichletAllocation
    • ColumnTransformer, FeatureUnion
    • OutputCodeClassifier, OneVsOneClassifier
    • HDBSCAN, FeatureAgglomeration, BisectingKMeans
    • GenericUnivariateSelect, SelectFdr, SelectFpr, SelectFwe, SelectKBest, SelectPercentile
    • HashingVectorizer, FeatureHasher, SparseRandomProjection, SkewedChi2Sampler
    • LocalOutlierFactor (predict-only)
  • Python function serialisation support (used by feature selection estimators)

[0.1.0-alpha.13] - 2024-10-15

Fixed

  • Dtype-robust sparse matrix comparison in tests
  • RandomTreesEmbedding re-enabled after OneHotEncoder fix

[0.1.0-alpha.12] - 2024-10-01

Added

  • Extended scikit-learn estimator support:
    • Birch, TunedThresholdClassifierCV
    • GradientBoostingClassifier, GradientBoostingRegressor
    • HistGradientBoostingClassifier, HistGradientBoostingRegressor
    • GaussianProcessClassifier, GaussianProcessRegressor (with kernel serialisation)
    • CalibratedClassifierCV, LinearDiscriminantAnalysis

[0.1.0-alpha.11] - 2024-09-15

Changed

  • Refactored serialization layer to a mixin-based architecture (NumpySerializerMixin, ScipySerializerMixin) for extensibility and modularity
  • Improved recursive deserialization for nested estimators and special types

[0.1.0-alpha.10] - 2024-09-01

Added

  • scikit-learn version tracking: the serialized payload now records the sklearn version used, and a UserWarning is raised on version mismatch at deserialization time
  • Dynamic TestPyPI badge in README

Fixed

  • CI badge auto-update loop

[0.1.0-alpha.5] - 2024-08-20

Added

  • Type and dtype tracking for model parameters during serialization
  • Support for nested estimators (e.g. pipelines, meta-estimators)
  • KDTree serialization support
  • IsotonicRegression, TweedieRegressor, PoissonRegressor, GammaRegressor support
  • NumPy array dtype preservation (fixes BaggingRegressor and similar)

Fixed

  • Serialization of numpy arrays of estimators
  • Pipeline serialization

[0.1.0-alpha.4] - 2024-08-15

Changed

  • Dynamic estimator loading using sklearn.utils.discovery.all_estimators
  • Improved attribute handling in SklearnSerializer

[0.1.0-alpha.1] - 2024-08-06

Added

  • Initial release of OpenModels library
  • Core functionality for serializing and deserializing machine learning models
  • Support for scikit-learn models:
    • Classification: LogisticRegression, RandomForestClassifier, SVC, BernoulliNB, GaussianNB, MultinomialNB, ComplementNB, Perceptron
    • Regression: LinearRegression, Lasso, Ridge, RandomForestRegressor, SVR
    • Clustering: KMeans
    • Dimensionality Reduction: PCA
    • Other: PLSRegression
  • JSON serialization format
  • Pickle serialization format
  • Extensible architecture for adding new model types and serialization formats
  • Basic test suite for supported models
  • Documentation including README, LICENSE, and CONTRIBUTING guidelines

Security

  • Implemented safe alternatives to pickle serialization

[Unreleased]

Planned

  • Support for TensorFlow models
  • YAML serialization format
  • Enhanced documentation with more examples and use cases
  • Support for more scikit-learn models including ensemble methods and neural networks