This is the release note of v1.2.0.

Highlights

Trial Queue

Study.enqueue_trial allows the user to specify fixed parameter values to be tried next instead of values from the suggestion algorithms. An example is available in the reference. See #520.

Note that this feature introduced an RDB schema change. If you have stored studies created by Optuna v1.1.0 or less in RDBs, please execute $ optuna storage upgrade --storage $URL after updating Optuna.

Experimental Grid Search

Grid search has been introduced as an experimental feature through GridSampler. See #665.

Breaking Changes

• Trial.report to not update the trial value. (#854)

New Features

• Add trial queue. (#520)
• Add a sampler class based on grid search. (#665)
• Display progress bar in optimization process. (#844)
• Trial.report to not update the trial value. (#854)
• Add experimental warning. (#884)
• experimental decorator to accept optional name and apply to progress bar. (#918)
• Add a n_startup_trials option to SkoptSampler. (#951)

Enhancements

• Fix OptunaSearchCV by adding the method that was implemented in scikit-learn>=0.22.1. (#881, thanks @himkt!)
• User-friendly error message for fixed categorical parameter not found in choices. (#900)
• Trial level suggest for same variable with different parameters give warning. (#908, thanks @PhilipMay!)
• More verbose warning message in Trial._check_distribution. (#934, thanks @PhilipMay!)

Bug Fixes

• Fix LightGBMTuner to handle metrics with evaluation positions. (#912)
• Fix a TPE implementation on DiscreteUniformDistribution. (#917)
  • If you use Trial.suggest_discrete_uniform with the default sampler (i.e., TPESampler), optimization performance may degrade due to the issue (#916). Please update Optuna to this version or later.
• Use weakref.finalize instead of __del__ for RDBStorage. (#941)

Documentation

• Add examples for doctest to optuna/trial.py. (#882, thanks @nuka137!)
• Add an example of enqueue_trial. (#927)
• Fix typo for cma.EvotionStrategy. (#929)
• Explain installing local Optuna explicitly. (#932, thanks @keisuke-umezawa!)
• Documentation of algorithm behind Optuna. (#940, thanks @arpitkh101!)
• Run a doctest of pycma sampler. (#944)
• Fix doctest in logging. (#953, thanks @keisuke-umezawa!)
• Fix doctest in parallel_coordinate. (#955, thanks @keisuke-umezawa!)
• Add examples for doctest to optuna/exceptions.py. (#958, thanks @nuka137!)
• Remove description and example of optuna.structs.TrialPruned (followup of #958). (#959)
• Add examples for doctest to optuna/prunes/*.py. (#964, thanks @nuka137!)
• Fix doctest in visualization. (#973, thanks @keisuke-umezawa!)

Examples

• Add an example of xgboost.cv using XGBoostPruningCallback. (#907, thanks @yutayamazaki!)
• TensorFlow estimator example with only v2 APIs. (#924, thanks @nuka137!)
• TensorFlow estimator integration example with only v2 APIs. (#935, thanks @nuka137!)
• Simplify PyTorch Ignite example by removing an evaluator. (#971)

Tests

• Add __init__.py to tests/samplers_tests/tpe_tests. (#945, thanks @keisuke-umezawa!)
• Cast to numpy.ndarray as Python lists are no longer accepted in xgboost==1.0. (#947)

Code Fixes

• Simplified _check_distribution. (#937, thanks @PhilipMay!)

Continuous Integration

• Remove version constraints of PyTorch and its related libraries. (#893)
• Bump up pip version. (#911)
• Fix sphinx version to avoid build error. (#942)
• Apply Python3-style of type hinting to optuna.integration.lightgbm_tuner.train. (#943)