keras-tuner

What's Changed

• Fixed issue #677 by @Anselmoo in https://github.com/keras-team/keras-tuner/pull/678
• Adopt safe model and trial saving practices in the multi-worker setting by @jamesmullenbach in https://github.com/keras-team/keras-tuner/pull/684
• tuner_utils: use datetime to calculate elapsed time by @mebeim in https://github.com/keras-team/keras-tuner/pull/690
• Add pre_create_trial callback by @jamesmullenbach in https://github.com/keras-team/keras-tuner/pull/695
• Multi-worker file writing checks by @jamesmullenbach in https://github.com/keras-team/keras-tuner/pull/694
• Update actions.yml by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/698
• Add "declare_hyperparameters" to HyperModel by @jamesmullenbach in https://github.com/keras-team/keras-tuner/pull/696
• Record best epoch info with update_trial by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/706

New Contributors

• @Anselmoo made their first contribution in https://github.com/keras-team/keras-tuner/pull/678
• @jamesmullenbach made their first contribution in https://github.com/keras-team/keras-tuner/pull/684
• @mebeim made their first contribution in https://github.com/keras-team/keras-tuner/pull/690

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.1.2...1.1.3rc0

What's Changed

• add --profile=black to isort by @LukeWood in https://github.com/keras-team/keras-tuner/pull/672
• In model checkpointing callback, check logs before get objective value by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/674

New Contributors

• @LukeWood made their first contribution in https://github.com/keras-team/keras-tuner/pull/672

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.1.1...1.1.2

What's Changed

• add --profile=black to isort by @LukeWood in https://github.com/keras-team/keras-tuner/pull/672
• In model checkpointing callback, check logs before get objective value by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/674

New Contributors

• @LukeWood made their first contribution in https://github.com/keras-team/keras-tuner/pull/672

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.1.1...1.1.2rc0

Highlights

• Support passing a list of objectives as the objective argument.
• Raise better error message when the return value of run_trial() or HyperModel.fit() are of wrong type.
• Various bug fixes for BayesianOptimization tuner.
• The trial IDs are changed from hex strings to integers counting from 0.

What's Changed

• Make hyperparameters names visible in Display output by @C-Pro in https://github.com/keras-team/keras-tuner/pull/634
• Replace import kerastuner with import keras_tuner by @ageron in https://github.com/keras-team/keras-tuner/pull/640
• Support multi-objective by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/641
• reorganize the tests to follow keras best practices by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/643
• keep Objective in oracle for backward compatibility by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/644
• better error check for returned eval results by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/646
• Mitigate the issue of hanging workers after chief already quits when running keras-tuner in distributed tuning mode. by @mtian29 in https://github.com/keras-team/keras-tuner/pull/645
• Ensure hallucination checks if the Gaussian regressor has been fit be… by @brydon in https://github.com/keras-team/keras-tuner/pull/650
• Resolves #609: Support for sklearn functions without sample_weight by @brydon in https://github.com/keras-team/keras-tuner/pull/651
• Resolves #652 and #605: Make human readable trial_id and sync trial numbers between worker Displays by @brydon in https://github.com/keras-team/keras-tuner/pull/653
• Update tuner.py by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/657
• fix(bayesian): scalar optimization result (#655) by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/662
• Generalize hallucination checks to avoid racing conditions by @alisterl in https://github.com/keras-team/keras-tuner/pull/664
• remove scipy from required dependency by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/665
• Import scipy.optimize by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/667

New Contributors

• @C-Pro made their first contribution in https://github.com/keras-team/keras-tuner/pull/634
• @ageron made their first contribution in https://github.com/keras-team/keras-tuner/pull/640
• @mtian29 made their first contribution in https://github.com/keras-team/keras-tuner/pull/645
• @brydon made their first contribution in https://github.com/keras-team/keras-tuner/pull/650
• @alisterl made their first contribution in https://github.com/keras-team/keras-tuner/pull/664

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.1.1rc0...1.1.1

Highlights

• Support passing a list of objectives as the objective argument.
• Raise better error message when the return value of run_trial() or HyperModel.fit() are of wrong type.

What's Changed

• Make hyperparameters names visible in Display output by @C-Pro in https://github.com/keras-team/keras-tuner/pull/634
• Replace import kerastuner with import keras_tuner by @ageron in https://github.com/keras-team/keras-tuner/pull/640
• Support multi-objective by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/641
• reorganize the tests to follow keras best practices by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/643
• keep Objective in oracle for backward compatibility by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/644
• better error check for returned eval results by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/646
• Mitigate the issue of hanging workers after chief already quits when running keras-tuner in distributed tuning mode. by @mtian29 in https://github.com/keras-team/keras-tuner/pull/645
• Ensure hallucination checks if the Gaussian regressor has been fit be… by @brydon in https://github.com/keras-team/keras-tuner/pull/650
• Resolves #609: Support for sklearn functions without sample_weight by @brydon in https://github.com/keras-team/keras-tuner/pull/651
• Resolves #652 and #605: Make human readable trial_id and sync trial numbers between worker Displays by @brydon in https://github.com/keras-team/keras-tuner/pull/653
• Update tuner.py by @haifeng-jin in https://github.com/keras-team/keras-tuner/pull/657

New Contributors

• @C-Pro made their first contribution in https://github.com/keras-team/keras-tuner/pull/634
• @ageron made their first contribution in https://github.com/keras-team/keras-tuner/pull/640
• @mtian29 made their first contribution in https://github.com/keras-team/keras-tuner/pull/645
• @brydon made their first contribution in https://github.com/keras-team/keras-tuner/pull/650

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.1.0...1.1.1rc0

What's Changed

• Support HyperModel.fit() to tune the fit process.
• Support Tuner.run_trial() to return a single float as the objective value to minimize.
• Support Tuner.run_trial() to return a dictionary of {metric_name: value} or Keras history.
• Allow not providing hypermodel to Tuner if override Tuner.run_trial().
• Allow not providing objective to Tuner if HyperModel.fit() or Tuner.run_trial() return a single float.
• Bug fixes

Breaking Changes

• Change internal class MultiExecutionTuner to Tuner to replace all overridden methods.
• Removed KerasHyperModel an internal class to wrap the user provided HyperModel.

New Contributors

• @liqiongyu made their first contribution in https://github.com/keras-team/keras-tuner/pull/594
• @vardhanaleti made their first contribution in https://github.com/keras-team/keras-tuner/pull/595
• @howl-anderson made their first contribution in https://github.com/keras-team/keras-tuner/pull/607

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.0.4...1.1.0rc0

What's Changed

• Support HyperModel.fit() to tune the fit process.
• Support Tuner.run_trial() to return a single float as the objective value to minimize.
• Support Tuner.run_trial() to return a dictionary of {metric_name: value} or Keras history.
• Allow not providing hypermodel to Tuner if override Tuner.run_trial().
• Allow not providing objective to Tuner if HyperModel.fit() or Tuner.run_trial() return a single float.
• Bug fixes

Breaking Changes

• Change internal class MultiExecutionTuner to Tuner to replace all overridden methods.
• Removed KerasHyperModel an internal class to wrap the user provided HyperModel.

New Contributors

• @liqiongyu made their first contribution in https://github.com/keras-team/keras-tuner/pull/594
• @vardhanaleti made their first contribution in https://github.com/keras-team/keras-tuner/pull/595
• @howl-anderson made their first contribution in https://github.com/keras-team/keras-tuner/pull/607

Full Changelog: https://github.com/keras-team/keras-tuner/compare/1.0.4...1.1.0rc0

• Support DataFrame in SklearnTuner.
• Support Tuner.search_space_summary() to print all the hyperparameters based on conditional_scopes.
• Support TensorFlow 2.0 for backward compatibility.
• Bug fixes and documentation improvements.
• Raise a warning when using with TF 1.
• Save TPUStrategy models with the TF format.

• Support DataFrame in SklearnTuner.
• Support Tuner.search_space_summary() to print all the hyperparameters based on conditional_scopes.
• Support TensorFlow 2.0 for backward compatibility.
• Bug fixes and documentation improvements.
• Raise a warning when using with TF 1.

• Support DataFrame in SklearnTuner.
• Support Tuner.search_space_summary() to print all the hyperparameters based on conditional_scopes.
• Support TensorFlow 2.0 for backward compatibility.
• Bug fixes and documentation improvements.

• Renamed import name of kerastuner to keras_tuner.
• Renamed the Oracles to add the Oracle as suffix, e.g., RandomSearch oracle is renamed to RandomSearchOracle. (The RandomSearch tuner is still named RandomSearch.)
• Renamed Tuner._populate_space to Tuner.populate_space.
• Renamed Tuner._score_trail to Tuner.score_trial.
• Renamed kt.tuners.Sklearn tuner to kt.SklearnTuner and put it at the root level of import.
• Removed the CloudLogger feature, but the Logger class still works.
• Tuning sklearn.pipeline.Pipeline.
• Improved the docstrings.

• Added Multi-worker DistributionStrategy support.
• Added EfficientNet application.
• Added application hypermodel with augmentation.
• Format console output information.
• Various bug fixes.

The first major release of Keras Tuner. This release includes improved algorithms and the ability to distribute tuning.