v0.2.0

Added

• Training option to ignore classes and features in less than N groups.
• cvi command run independent cross-validation as done in the paper.
• Support for training and evaluating random forest models.
• Support for computing sample weights for observation groups in fit.
• Report information content of the prediction in predict output.

Changed

• Relax anndata dependency to allow more recent releases.
• Update MIBiG 3.1 dataset to filter out erroneous partial clusters.
• Retrain CHAMOIS with filtered MIBIG 3.1 dataset and improved feature selection.
• Make chamois.compositions.build_variables drop empty columns.
• Rename screen command to compare.
• Incorrect use of chemical groups in cross-validation commands.
• Update Pfam to 38.0.

Fixed

• Issue with computation of probabilistic Jaccard distance in validate command.
• Incorrect error message on non-existing class in explain command.
• Incorrect merging of multiple feature tables in CLI.
• Extraction of minimum class and features by groups for majority positive classes.
• Incorrect error message in predict command.
• Unsupported sample weighting in older scikit-learn releases.