UTest

class hypex.comparators.UTest(compare_by, grouping_role=None, target_role=None, baseline_role=None, reliability=0.05, key='', calc_kwargs={})[source]

Bases: StatHypothesisTesting

Mann-Whitney U test (Wilcoxon rank-sum) for numeric targets.

Non-parametric alternative to the t-test; compares rank distributions rather than means. Operates on raw data slices via scipy.

Parameters:

  • compare_by (Literal['groups', 'columns', 'columns_in_groups', 'cross', 'matched_pairs']) – Comparison mode ("groups", "columns", etc.).

  • grouping_role (Optional[ABCRole]) – Role that identifies the group column.

  • target_role (Optional[ABCRole]) – Role that identifies the numeric target column(s).

  • reliability (float) – Significance level α (default 0.05).

  • key (Any) – Optional label for this test instance.