utils Package

Various useful constants defining groups of evaluation metrics.

skll.utils.constants.CLASSIFICATION_ONLY_METRICS = {'accuracy', 'average_precision', 'balanced_accuracy', 'f05', 'f05_score_macro', 'f05_score_micro', 'f05_score_weighted', 'f1', 'f1_score_least_frequent', 'f1_score_macro', 'f1_score_micro', 'f1_score_weighted', 'jaccard', 'jaccard_macro', 'jaccard_micro', 'jaccard_weighted', 'neg_log_loss', 'precision', 'precision_macro', 'precision_micro', 'precision_weighted', 'recall', 'recall_macro', 'recall_micro', 'recall_weighted', 'roc_auc'}

Set of evaluation metrics only used for classification tasks

skll.utils.constants.CORRELATION_METRICS = {'kendall_tau', 'pearson', 'spearman'}

Set of evaluation metrics based on correlation

skll.utils.constants.PROBABILISTIC_METRICS = frozenset({'average_precision', 'neg_log_loss', 'roc_auc'})

Set of evaluation metrics that can use prediction probabilities

skll.utils.constants.REGRESSION_ONLY_METRICS = {'explained_variance', 'max_error', 'neg_mean_absolute_error', 'neg_mean_squared_error', 'neg_root_mean_squared_error', 'r2'}

Set of evaluation metrics only used for regression tasks

skll.utils.constants.UNWEIGHTED_KAPPA_METRICS = {'unweighted_kappa', 'uwk_off_by_one'}

Set of unweighted kappa agreement metrics

skll.utils.constants.WEIGHTED_KAPPA_METRICS = {'linear_weighted_kappa', 'lwk_off_by_one', 'quadratic_weighted_kappa', 'qwk_off_by_one'}

Set of weighed kappa agreement metrics

A useful logging function for SKLL developers

skll.utils.logging.get_skll_logger(name, filepath=None, log_level=20)[source]

Create and return logger instances appropriate for use in SKLL code.

These logger instances can log to both STDERR as well as a file. This function will try to reuse any previously created logger based on the given name and filepath.

  • name (str) – The name to be used for the logger.

  • filepath (Optional[str], default=None) – The file to be used for the logger via a FileHandler. Default: None in which case no file is attached to the logger.

  • log_level (int, default=logging.INFO) – The level for logging messages


logger – A Logger instance.

Return type: