This release includes almost all open PR from the original repository since 2020. The original scikit-optimize repo is now archived, I will continue development on https://github.com/holgern/scikit-optimize. Please add issues and PRs here.
- Feature Add support for multimetric scoring to :obj:
skopt.searchcv.BayesSearchCV. - Feature Evaluate random point instead of using point twice in :obj:
skopt.optimizer.Optimizer. - Feature :obj:
skopt.callback.CheckpointSavercan now recycle previous function evaluations - Feature finish CI move to GHA
- Fix Keep order of variables in LabelEncoder
- Feature add from_df method to Space class
- Fix Fix :obj:
skopt.plots.plot_evaluationsincorrectly handling dimensions
as categorical dimensions when constant dimension are ignored due to wrong
indexing. - Fix Improved Bayesian Optimization documentation
- Fix Replace occurrences of mse with squared_error
- Fix numpy.int issue
- Feature add standard deviation Stopper for Gaussian process
- Feature Implement MES and PVRS acquisition functions.
- Fix Fix tuple index out of range on plot
- Fix Fixes Grid sampling with border="only"
- Fix Update plot_evaluations param
- Fix Fix plot_gaussian_process not working with ps-acquisition
- Feature Make Real and Integer raise error when prior is log-uniform and bounds contain zero
- Fix Better use proper parsing of the scikit-learn version numbers.
- Feature Test DeltaYStopper with minimize functions fixes
- Fix dimension mismatch with gp_minimize in Matern kernel when nu=0.5
- Fix GPR with noise fix
- Feature Add constrained optimization
- Feature Add 'ax' argument to all plotting functions
- API Improve and tighten the inference of
skopt.space.space.Dimension
objects fromtuple,dictand numpy array. See the documentation of
skopt.space.space.check_dimensionfor details. The old behavior is
retained for a transitionary period; a warning is raised if the inference
would change under the new scheme. - Minor documentation improvements.
- Various small bugs and fixes.