v0.7.2

• MAINT #671465e2: Make loading json files from earlier versions backwards compatible

v0.7.1

ci: bump version to 0.7.1

Version 0.6.1

• MAINT #286: Add support for Python 3.11.
• FIX #282: Fixes a memory leak in the neighborhood generation of integer hyperparameters.

v0.6.0

v0.6.0

• ADD #255: An easy interface of Float, Integer, Categorical for creating search spaces. Update documentation
• ADD #243: Add forbidden relations between two hyperparamters
• MAINT #243: Change branch master to main
• FIX #259: Numpy runtime error when rounding
• FIX #247: No longer errors when serliazing spaces with an InCondition
• FIX #219: Hyperparamters correctly active with diamond-or conditions

v0.5.0

• Fix #231: Links to the pcs formats.
• Fix #230: Allow Forbidden Clauses with non-numeric values.
• Fix #232: Equality == between hyperparameters now considers default values.
• Fix #221: Normal Hyperparameters should now properly sample from correct distribution in log space
• Fix #221: Fixed boundary problems with integer hyperparameters due to numerical rounding after sampling.
• Maint #221: Categorical Hyperparameters now always have associated probabilities, remaining uniform if non are provided. (Same behaviour)
• Add #222: BetaFloat and BetaInteger hyperparamters, hyperparameters distributed according to a beta distribution.
• Add #241: Implements support for PiBo, you can now embed some prior distribution knowledge into ConfigSpace hyperparameters.
  • See the example here.
  • Hyperparameters now have a pdf(vector: np.ndarray) -> np.ndarray to get the probability density values for the input
  • Hyperparameters now have a get_max_density() -> float to get the greatest value in it's probability distribution function, the probability of the mode of the distriubtion.
  • ConfigurationSpace objects now have a remove_parameter_priors() -> ConfigurationSpace to remove any priors

v0.4.21

Version 0.4.21

• Add #224: Now builds binary wheels for Windows/Mac/Linux, available on PyPI.
• Maint #227: Include automated testing for windows and mac.
• Maint #228: #226: Account for test differences with i686 architectures.
• Maint #213, #215: Prevent double trigger of github workflows.
• Fix #212: Equality (==) on CategoricalHyperparameter objects are now invariant to ordering.
• Add #208: ConfigurationSpace::estimate_size() to get the size of a configuration space without considering constraints.
• Add #210: print(config) is now produces a string representation of a valid python dictionary that is suitable for copy and paste.
• Fix #203: Parser for pcs files now correctly coverts types for forbidden clauses, checking for the validaty as well.
• Maint #f71508c: Clean up in README.md and fix link for new SMAC example docs.
• Fix #202: Fix numerical underflow when performing quantization of log sampled UniformFloat.
• Add #188: Support for a truncated NormalIntegerHyperparameter or NormalFloatHyperparameter by providing lower and upper bounds.
• Fix #195: Sampling configurations to perform validity checks for during get_one_exchange_neighborhood is now deterministic w.r.t. a seed.

v0.4.20

Version 0.4.20

• MAINT #185: Drop support for Python 3.6
• FIX #190: Remove old files with old GPL-3.0 license
• ADD #191: Configuration and ConfigurationSpace can now act as mappings

Version 0.4.19

• ADD #184: Wheels.
• FIX #176: copy meta field in add_configuration_space
• MAINT #181: Run Flake8 on Cython code
• MAINT #182: Replace rich comparisons by __eq__ in Cython code
• MAINT #183: Cleanup warnings.

Version 0.4.18

Version 0.4.18

• ADD #164: New method rvs for hyperparameters to allow them being used with scikit-learn's hyperparameter optimization tools.
• FIX #173: Fixes a numpy ABI incompatibility problem with numpy 1.20

Version 0.4.17

• MAINT #168: Support for Python. 3.9.X