Changes:
- Fix saving and loading DGP graphs from persistent storage (#151)
- Add Python tests (#149)
- Drop support for Python 3.8; Add support for Python 3.13 (#155)
Changes:
- Update Python packaging (#124)
- Add Python
validation_data
input check (#136) - Fix sanitizer, free test memory, fix neural saving (#146)
- Update cJSON and pybind11 libs (#147)
Changes:
- Remove compilation with
-flto
(#128)
Changes:
- Improve parameter checks (#112, #114)
- Improve hyperrectangle CSR semantics (#117)
- Fix a sampling bug in Python
fit()
whenshuffle=False
(#122)
Changes:
- Make predict function deterministic even with parallelism (#109)
Changes:
Changes:
- Make
random_state
parameter sklearn compatible (#97)
Changes:
- Add tests.
- Fix a bug serialising initial population set filename.
- Fix a bug copying convolutional layer momentum.
Changes:
- Accept 1-D flattened Python inputs in fit/score/predict (#81)
- Fix setting max trials via JSON (#83)
- Remove Python
seed()
and addRANDOM_STATE
parameter for setting seed (#86) - Major Python API update: sklearn compatibility (#77)
- Add Python callback support with EarlyStoppingCallback (#77)
- Add parameter to read initial population from JSON; also works in stand-alone binary now (#77)
- Add hyperparameter tuning example (#77)
- Add CheckpointCallback (#88)
Changes:
Changes:
- Add support for pickle to Python library (#72)
Changes:
- Add defaults to Python API calls to allow a cleaner syntax (#60)
- Fix an input check on the cover parameter in predict/score (#60)
- Update Python cartpole example to latest version and refresh other examples (#60)
Changes:
- Add an optional argument to the Python predict/score functions that specifies the value to return for a sample if the match set is empty instead of invoking covering (#59)
Changes:
- Add docstrings and variable names to Python library (#43)
- Fix minor memory leak when printing parameters (#44)
Changes:
- Python library throws exceptions rather than hard exiting where possible (#41, #42)
- Fix minor memory leak when printing parameters.
Changes:
- Fix numpy subnormal warning.
Changes:
- Best action selection now breaks ties randomly (#39)
Changes:
- Fix hyperrectangle_ubr json population seeding import/export asymmetry.
Changes:
- Added extra JSON parsing input checks.
- Cleaned up Python examples.
- Added Jupyter notebook examples.
- Renamed hyperrectangle conditions with hyperrectangle_csr (#35)
- Added unordered-bound hyperrectangle conditions with hyperrectangle_ubr (#35)
Changes:
- Fixed cross-platform compiling and wheel building.
Changes:
- Added functions to import parameters as JSON.
- Stand-alone binary config file now in JSON format.
- Python library and stand-alone config parameter setters use same JSON import functions.
- Added functions inserting classifiers into the population in JSON.
- Minor refactoring.
Changes:
- Python utilities included in packaging.
- Python code formatted with black.
- Added Python type hints and stub for mypy type checking.
- Hyphens in variable names/types converted to underscore.
Changes:
- Added cJSON library.
- Added functions returning/printing classifiers in JSON.
- Added functions returning/printing parameters in JSON.
- Added Python classes for visualising tree and graph conditions.
- Fixed Python read/write
EA_SELECT_TYPE
asymmetry (#16) - Cleaned up Python interface and added input checks.
- Moved ini config to a subdirectory.
- Main executable building now optional.
- Added pypi package building and support building Windows Python 3.10.
- Updated libraries to latest versions.
Changes:
- Clean up - move to wiki.
- Addition of set seed function (#14)
- Minor bug fix to action/match set numerosity.
- Neural layers print more detailed information.
Changes:
- Fixed CMakeLists to remove schematic warning.
Changes:
- Increased documentation.
- Additional unit tests.
- Float-to-binary changed to human readable ordering.
- Fixed opening config ini file on Windows.
- Minor source refactoring/cleaning.
- Added neural layer initialisation parameters to file saving/loading.
First version.