23.0.0.post1: Fix scikit-learn dependency

This version specifies an upper limit on the installed scikit-learn version.
GAMA is incompatible with versions >=1.3, as it builds on scikit-learn internals that changed.

Full Changelog: v23.0.0...v23.0.0.post1

v23.0.0

What's Changed

• Updating devtools, removing dashboard by @PGijsbers in #192
• Bump to minimum Py3.9 by @PGijsbers in #195
• Refactor by @PGijsbers in #196
• Add back preset hyperparameter (was goal in 21.0.1) by @PGijsbers in #203

New Contributors

• @darigovresearch Added license to readme by @darigovresearch in #187

Full Changelog: v22.0.0...v23.0.0

GAMA 22.0.0

What's Changed

Bugfixes:

• #137: raise an output if output_directory is non-empty.
• #174: Fix an issue where GAMA might freeze in when ending search.

Features:

• ASHA resources can now also be specified as fraction of the dataset size by using a float in (0, 1].

Changes:

• #138: Instead of subsampling the whole dataset before doing CV, the same test set is used across rungs and only
  the training folds are subsampled. This makes performance comparable across rungs.
• AsyncEA will try to delay creating new individuals as long as possible.
• AsyncEA will no longer create offspring from pipelines with failed evaluations (e.g., timeout, invalid configuration).

Maintenance:

• Adopt NEP 29:
  • Drop support for Python 3.6 and 3.7
  • Add support for Python 3.10
• Use pyproject.toml instead of setup.py
• Transition to Github CI, including:
  • pytest on pull request
  • pytest CRON job with prerelease versions
  • doc build and deploy
  • pre-commit check
  • check for changelog
  • easy release to pypi from github workflow
• Small changes to avoid FutureWarnings and/or DeprecationWarnings.

New Contributors

• @himanshu007-creator made their first contribution in #118
• @chclam made their first contribution in #148

Full Changelog: 21.0.0...v22.0.0

20.0.1

Add a goal hyperparameter that can be used to let GAMA configure the AutoML hyperparameter based on user intent.

21.0.0 Release

Version 21.0.0

Features:

• The warm_start parameter of fit() had a slight behavior change (see docs).
• Fit can now be called more than once. It will use the same time constraint as the first fit call.

Bugfixes:

• TargetEncoder is no longer used for Classification tasks, since there is a bug that prevents it from working as intended.

Version 20.2.1

Changes:

• # 24: Changes to logging
  The structure of the log file(s) have changed.
  The goal is to make the log files easier to use, by making them easier to read and
  easier to extend write behavior.
  There will now be three log files, one which contains just evaluation data, one which contains progress data, and one which contains resource usage data.
  For more information see :ref:logging-section in the technical guide.

Features:

• # 66: csv files are now supported.
  Call fit_arff is now fit_from_file which accepts both arff and csv files.
  The CLI interface and Gama Dashboard also allow for csv files.
• # 92: You can specify a memory limit through max_memory_mb hyperparameter.
  GAMA does not guarantee it will not violate the constraint, but violations
  should be infrequent and minor. Feel free to open an issue if you experience a
  violation which does not minor.

GAMA v20.2.0

Features:

70: Exported code is now auto-formatted with Black.

• The Dashboard now has an Analysis tab that allows you to load old log files.
• The Dashboard Home tab allows you to view the data set.
• The Dashboard Home tab allows you to select a target column.
• Pipelines and estimators are now cached on disk again.
• KNN and PolynomialFeatures are now dynamically disabled based on dataset size.

Maintenance:

• Docs are automatically built and deployed on a commit to master and develop.
• Pre-commit configuration added to check formatting, style and type hints on commit.
• Black codestyle adapted, most drastic change is line length from 120 to 88.
• Coverage increased by removing unused code, updating configuration, adding tests.
• Memory usage of all GAMA's processes is logged.

GAMA v19.11.2

Bugfixes:

• predict_proba_arff now also accepts a target_column as expected from the previous update.

GAMA v19.11.1

New since 19.8.0

Features:

• gama.__version__ can now be used to retrieve gama's version.
• fit_arff, score_arff and predict_arff now accept a target_column parameter to specify the target.
  If left unset, the last column of the ARFF file is assumed to be the target column.

Bugfixes:

• fit(x, y) may now be called with y as (N,1) array.
• ensemble post-processing is now compatible with non-zero indexed class labels

Maintenance:

• __version__.py is now the only place with hard-coded version.

First PyPI release

First PyPI release. Includes:

• GamaClassifier and Regressor
• Configurable AutoML pipeline with multiple search methods and post processing steps
• WIP Log visualization

Would love feedback