[2.0.0] - 2025-10-11

Changed

• BREAKING: Removed scikit-learn dependency and GeneticSearchCV class
• BREAKING: Minimum Python version increased from 3.8 to 3.10
• Migrated from setuptools to hatchling build backend
• Migrated from pip to uv for dependency management
• Modernized CI/CD workflows to use uv
• Updated README with modern development workflow

Added

• Comprehensive type hints across entire codebase
• NumPy-style docstrings on all public APIs
• Makefile with helpful development targets (test, lint, format, typecheck, build, coverage, clean)
• MyPy type checking with strict mode enabled
• Ruff for linting and formatting with docstring validation
• Coverage reporting configuration
• Python 3.13 and 3.14 support
• Tournament selection option for improved parent selection pressure (use_tournament_selection, tournament_size parameters)
• Diversity protection with automatic population diversity monitoring and injection when diversity falls below threshold
  • Handles both numeric parameters (variance-based) and categorical choice parameters (unique ratio)
  • Automatically injects random individuals when diversity falls below threshold
• Adaptive mutation that automatically adjusts exploration vs exploitation based on optimization progress
• Early stopping via patience parameter to halt optimization when no improvement is seen for N generations
• Enhanced mutation with mutation_rate parameter (0.0-1.0) for probabilistic per-parameter mutation
• Comprehensive test suite with 15 new tests covering all new features (64 total tests, 97% coverage)

Fixed

• ReadTheDocs configuration for proper documentation building
• Docstring formatting for Sphinx compatibility
• Infinite loop bug in breeding logic where if set1 != set2 condition could hang indefinitely
• Incomplete random seeding - now seeds both random and numpy.random for full reproducibility
• Weak elitism protection - top performers are now preserved correctly across generations
• Improved breeding algorithm to always produce valid offspring

What's Changed

• Add Python 3.11 support
• Remove Python 3.7 support

Full Changelog: v1.3.0...v.1.4.0

New Feature

• Added the ability to provide an asynchronous loss/reward function to the optimize function.

Miscellaneous Changes

• Added documentation to https://geneticpy.readthedocs.io.

Full Changelog: v1.2.3...v1.3.0

Bug Fix

• Updated Scikit-Learn support to include >=0.23.2.

Miscellaneous Changes

• Changed CI/CD from Travis to GitHub Actions
• Utilizing pyproject.toml and setup.cfg files instead of setup.py
• Using setuptools-git-versioning for automatic versioning with release.
• Adding Release step to GitHub actions to automate deployment to Pypi.

Bug Fix

• Fixed issue with complex pipelines and GeneticSearchCV.

Bug Fix

• Complex pipelines can now be used with GeneticSearchCV.

New Features

• Introduced GeneticSearchCV class.

Changes

• Removed Python 3.4 and Python 3.5 support.

Bug Fixes

• Using tqdm autonotebook to allow tqdm counter to automatically detect Jupyter notebooks.
• Fixed incorrect counting of tqdm counter.

Other Changes

• Returning dictionary by default for the optimize function.
• Added DeprecationWarning for "tuple" optimize return type.

New Features

• Created dictionary return type for optimize function.

Other Changes

• Added PendingDeprecationWarning for "tuple" optimize return type.
• Added Python 3.8 testing.

New Features

• Created 'seed' parameter to allow for reproducible results.
• Allowing for early stopping when 'target' loss is reached.
• Added 'total_time' to output of the optimize() function.

Bug Fixes

• Exception is raised if loss function returns None.

Other Changes

• Added test coverage report.
• Achieved 100% test coverage.