fix(deps): update dependency keras-tuner to v1.4.7

[renovate]

This PR contains the following updates:

Package	Change	Age	Confidence
keras-tuner	==1.1.3 -> ==1.4.7

---

Release Notes

keras-team/keras-tuner (keras-tuner)

v1.4.7

Compare Source

Bug fixes

• Changed the chief server waiting time before shutting down to 60 minutes
  by default.

New Contributors

• @​sfo made their first contribution in #​977
• @​onponomarev made their first contribution in #​991

Full Changelog: keras-team/keras-tuner@v1.4.6...v1.4.7

v1.4.6

Compare Source

Bug fixes

• When running in parallel, the chief may exit before some client ask for another trial, which informs the client to exit. Now, it is fixed.

New features

• Updated the dependency from keras-core to keras version 3 and above. Also support keras version 2 for backward compatibility.

New Contributors

• @​AniketP04 made their first contribution in #​962

Full Changelog: keras-team/keras-tuner@v1.4.5...v1.4.6

v1.4.5

Compare Source

Bug fixes

• When running in parallel, the client oracle used to wait forever when the chief oracle is not responding. Now, it is fixed.
• When running in parallel, the client would call the chief after calling oracle.end_trial(), when the chief have already ended. Now, it is fixed.
• When running in parallel, the chief used to start to block in tuner.__init__(). However, it makes more sense to block when calling tuner.search(). Now, it is fixed.
• Could not do from keras_tuner.engine.hypermodel import HyperModel. It is now fixed.
• Could not do from keras_tuner.engine.hyperparameters import HyperParameters. It is now fixed.
• Could not do from keras_tuner.engine.metrics_tracking import infer_metric_direction. It is now fixed.
• Could not do from keras_tuner.engine.oracle import Objective. It is now fixed.
• Could not do from keras_tuner.engine.oracle import Oracle. It is now fixed.

Full Changelog: keras-team/keras-tuner@v1.4.4...v1.4.5

v1.4.4

Compare Source

Bug fixes

• Could not do from keras_tuner.engine.hyperparameters import serialize. It is now fixed.
• Could not do from keras_tuner.engine.hyperparameters import deserialize. It is now fixed.
• Could not do from keras_tuner.engine.tuner import maybe_distribute. It is now fixed.

Full Changelog: keras-team/keras-tuner@v1.4.3...v1.4.4

v1.4.3

Compare Source

Bug fixes

• Could not do from keras_tuner.engine.tuner import Tuner. It is now fixed.
• When TensorFlow version is low, it would error out with keras models have no
  attributed called get_build_config. It is now fixed.

Full Changelog: keras-team/keras-tuner@v1.4.2...v1.4.3

v1.4.2

Compare Source

Bug fixes

• Could not do from keras_tuner.engine import trial. It is now fixed.

Full Changelog: keras-team/keras-tuner@v1.4.1...v1.4.2

v1.4.1

Compare Source

Bug fixes

• Could not do from keras_tuner.engine import base_tuner. It is now fixed.

Full Changelog: keras-team/keras-tuner@v1.4.0...v1.4.1

v1.4.0

Compare Source

Breaking changes

• All private APIs are hidden under keras_tuner.src.*. For example, if you use
  keras_tuner.some_private_api, it will now be
  keras_tuner.src.some_private_api.

New features

• Support Keras Core with multi-backend.

New Contributors

• @​airvzxf made their first contribution in #​906
• @​pnacht made their first contribution in #​930
• @​fhausmann made their first contribution in #​926

Full Changelog: keras-team/keras-tuner@v1.3.5...v1.4.0rc

v1.3.5

Compare Source

Breaking changes

• Removed TensorFlow from the required dependencies of KerasTuner. The user need
  to install TensorFlow either separately with KerasTuner or with
  pip install keras_tuner[tensorflow]. This change is because some people may
  want to use KerasTuner with tensorflow-cpu instead of tensorflow.

Bug fixes

• KerasTuner used to require protobuf version to be under 3.20. The limit is
  removed. Now, it support both protobuf 3 and 4.

Full Changelog: keras-team/keras-tuner@v1.3.4...v1.3.5

v1.3.4

Compare Source

Bug fixes

• If you have a protobuf version > 3.20, it would through an error when import KerasTuner. It is now fixed.

v1.3.3

Compare Source

• KerasTuner would install protobuf 3.19 with protobuf<=3.20. We want to install 3.20.3, so we changed it to protobuf<=3.20.3. It is now fixed.

Full Changelog: keras-team/keras-tuner@v1.3.2...v1.3.3

v1.3.2

Compare Source

Bug fixes

• It use to install protobuf 4.22.1 if install with TensorFlow 2.12, which is not compatible with KerasTuner. We limited the protobuf version to <=3.20, which is compatible with all TensorFlow versions so far.

Full Changelog: keras-team/keras-tuner@v1.3.1...v1.3.2

v1.3.1

Compare Source

Bug fixes

• The Tuner.results_summary() did not print error messages for failed trials and did not display Objective information correctly. It is now fixed.
• The BayesianOptimization would break when not specifying the num_initial_points and overriding .run_trial(). It is now fixed.
• TensorFlow 2.12 would break because the different protobuf version. It is now fixed.

New Contributors

• @​jkittner made their first contribution in #​860

Full Changelog: keras-team/keras-tuner@v1.3.0...v1.3.1

v1.3.0

Compare Source

Breaking changes

• Removed Logger and CloudLogger and the related arguments in BaseTuner.__init__(logger=...).
• Removed keras_tuner.oracles.BayesianOptimization, keras_tuner.oracles.Hyperband, keras_tuner.oracles.RandomSearch, which were actually Oracles instead of Tuners. Please usekeras_tuner.oracles.BayesianOptimizationOracle, keras_tuner.oracles.HyperbandOracle, keras_tuner.oracles.RandomSearchOracle instead.
• Removed keras_tuner.Sklearn. Please use keras_tuner.SklearnTuner instead.

New features

• keras_tuner.oracles.GridSearchOracle is now available as a standalone Oracle to be used with custom tuners.

Full Changelog: keras-team/keras-tuner@1.2.1...v1.3.0

v1.2.1

Compare Source

Bug fixes

• The resume feature (overwrite=False) would crash in 1.2.0. This is now fixed.

New Contributors

• @​nkovela1 made their first contribution in #​834

Full Changelog: keras-team/keras-tuner@1.2.0...1.2.1

v1.2.0

Compare Source

Release v1.2.0

Breaking changes

• If you implemented your own Tuner, the old use case of reporting results with Oracle.update_trial() in Tuner.run_trial() is deprecated. Please return the metrics in Tuner.run_trial() instead.
• If you implemented your own Oracle and overrided Oracle.end_trial(), you need to change the signature of the function from Oracle.end_trial(trial.trial_id, trial.status) to Oracle.end_trial(trial).
• The default value of the step argument in keras_tuner.HyperParameters.Int() is changed to None, which was 1 before. No change in default behavior.
• The default value of the sampling argument in keras_tuner.HyperParameters.Int() is changed to "linear", which was None
  before. No change in default behavior.
• The default value of the sampling argument in keras_tuner.HyperParameters.Float() is changed to "linear", which was
  None before. No change in default behavior.
• If you explicitly rely on protobuf values, the new protobuf bug fix may affect you.
• Changed the mechanism of how a random sample is drawn for a hyperparameter. They now all start from a random value between 0 and 1, and convert the value to a random sample.

New features

• A new tuner is added, keras_tuner.GridSearch, which can exhaust all the possible hyperparameter combinations.
• Better fault tolerance during the search. Added two new arguments to Tuner and Oracle initializers, max_retries_per_trial and max_consecutive_failed_trials.
• You can now mark a Trial as failed by raise keras_tuner.FailedTrialError("error message.") in HyperModel.build(), HyperModel.fit(), or your model build function.
• Provides better error messages for invalid configs for Int and Float type hyperparameters.
• A decorator @keras_tuner.synchronized is added to decorate the methods in Oracle and its subclasses to synchronize the concurrent calls to ensure thread safety in parallel tuning.

Bug fixes

• Protobuf was not converting Boolean type hyperparameter correctly. This is now fixed.
• Hyperband was not loading the weights correctly for half-trained models. This is now fixed.
• KeyError may occur if using hp.conditional_scope(), or the parent argument for hyperparameters. This is now fixed.
• num_initial_points of the BayesianOptimization should defaults to 3 * dimension, but it defaults to 2. This is now fixed.
• It would through an error when using a concrete Keras optimizer object to override the HyperModel compile arg. This is now fixed.
• Workers might crash due to Oracle reloading when running in parallel. This is now fixed.

New Contributors

• @​Firas-RHIMI made their first contribution in #​711
• @​HanxiaoLyu made their first contribution in #​746
• @​leleogere made their first contribution in #​794
• @​LuNoX made their first contribution in #​815

Full Changelog: keras-team/keras-tuner@1.1.3...1.2.0

---

Configuration

📅 Schedule: Branch creation - At any time (no schedule defined), Automerge - At any time (no schedule defined).

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.