Version 2.3.0

• Change in library support

  • Dis-continue python 3.7 support, but add 3.10 support.

  • Updated scikit-learn requiriment to >= 1.1.0, requiring the above change

  • Bump the nevergrad version to more stable 5.0

  • Require threadpoolctl>=3

• Update model param defaults

  • Updated the default regression pre-set model params for ridge regression, random forest and lgbm.

  • These new params are based off extensive benchmarking.

• NaN support in Loader

  • Loaders with will now support propegating any found NaN’ with the same shape as the rest of the loaded data files.

  • Will support ‘single’ and ‘all’ behavior, tests added to confirm.

• Added Dataset.display_scopes

  • Allows viz HTML repr of dataset by scope instead of default role.

• Cosmetic change from role name ‘data’ to ‘input data’

  • All methods will still work when passed ‘data’ indefinetly, but internally, and for display, switch to having the role key as ‘input data’ instead.

• Fix bug with permutation testing

  • Block structure and custom scopes will now work together better.

• Fix bug in Dataset.update_data_file_paths()

  • Now handles NaN case.

• Tweaks to input parameters

  • Just allow more options when init’ing CV, ProblemSpec and ParamSearch.

• Removed MLP

  • Why are you using MLP? Don’t, it just isn’t good. Sorry, gone, poof.

• New experimental color options

  • Messing around with option for coloring output for Compare summary, but still broken for now.