This repository has been archived by the owner on Feb 28, 2024. It is now read-only.
Releases: scikit-optimize/scikit-optimize
Releases · scikit-optimize/scikit-optimize
v0.7rc2
v0.7rc1
Highly composite six - v0.6
Version 0.6
Highly composite six.
New features
plot_regret
function for plotting the cumulative regret;
The purpose of such plot is to access how much an optimizer
is effective at picking good points.CheckpointSaver
that can be used to save a
checkpoint after each iteration with skopt.dumpSpace.from_yaml()
to allow for external file to define Space parameters
Bug fixes
- Fixed numpy broadcasting issues in gaussian_ei, gaussian_pi
- Fixed build with newest scikit-learn
- Use native python types inside BayesSearchCV
- Include fit_params in BayesSearchCV refit
Maintenance
- Added
versioneer
support, to reduce changes with new version of theskopt
v0.6rc1
Release candidate for v0.6.
v0.5.2
Version 0.5.2
Bug fixes
- Separated
n_points
fromn_jobs
inBayesSearchCV
. - Dimensions now support boolean np.arrays.
Maintenance
matplotlib
is now an optional requirement (install withpip install 'scikit-optimize[plots]'
)
v0.5.1 - re-release
This release super-seeds v0.5. We made a mistake tagging the v0.5rc1 on PyPI and releasing this to replace v0.5.
High five - v0.5
Version 0.5
High five!
New features
- Single element dimension definition, which can be used to fix the
value of a dimension during optimization. total_iterations
property ofBayesSearchCV
that counts total
iterations needed to explore all subspaces.- Add iteration event handler for
BayesSearchCV
, useful for early
stopping insideBayesSearchCV
search loop. - Added
utils.use_named_args
decorator to help with unpacking
named dimensions when calling an objective function.
Bug fixes
- Removed redundant estimator fitting inside
BayesSearchCV
. - Fixed the log10 transform for Real dimensions that would lead
to values being out of bounds.
v0.5rc1
Release candidate for v0.5.
Go forth - v0.4
Version 0.4
Go forth!
New features
- Support early stopping of optimization loop.
- Benchmarking scripts to evaluate performance of different surrogate models.
- Support for parallel evaluations of the objective function via several
constant liar stategies. - BayesSearchCV as a drop in replacement for scikit-learn's GridSearchCV.
- New acquisition functions "EIps" and "PIps" that takes into account
function compute time.
Bug fixes
- Fixed inference of dimensions of type Real.
API changes
- Change interface of GradientBoostingQuantileRegressor's predict method to
match return type of other regressors - Dimensions of type Real are now inclusive of upper bound.