Change Logs#

v0.8.1 - Mar 8, 2023#

Fix several bugs in transfer learning.
Fix several bugs & improve usability of HTMLVisualizer (#58).
Update time limit function to support multi-platform and multi-version of python. Update API. (#59)

Optimizer: runtime_limit -> max_runtime. time_limit_per_trial -> max_runtime_per_trial (#59).

Add HTML visualization for the optimization process (#48).
- Provide basic charts for objectives and constraints.
- Provide advanced functions, including surrogate fitting analysis and hyperparameter importance analysis.
Update transfer learning (#54).
- API change: for transfer learning data, user should provide a List[History] as transfer_learning_history, instead of a OrderedDict[config, perf] as history_bo_data (#54, 4641d7cf).
- Examples and docs are updated.
Refactor History object (0bce5800).
- Rename HistoryContainer to History.
- Simplify data structure and provide convenient APIs.
- Rewrite all methods, including data obtaining, plotting, saving/loading, etc.

API change: objs are renamed to objectives. num_objs are renamed to num_objectives (ecd5928a).
Change objective value of failed trials from MAXINT to np.inf (da88bd24).
Drop support for Python 3.6 (end of life on Dec 23, 2021).

Add BlendSearch, LineBO and SafeOpt (experimental) (#40).
Add color logger. Provide fine-grained control of logging options (e.g., log level).
Rewrite python packaging of the project (#55).
Update Markdown parser in docs to myst-parser. recommonmark is deprecated.
Add pytest for examples.
Use GitHub Actions for CI/CD.

Fix error return type of generic advisor and update sampler (Thanks @yezoli) (#44).
Consider constraints in plot_convergence (#47).

Add ConditionedSpace to support complex conditions between hyperparameters (#37).
Numerous bug fixes.

Project started.