API Reference / agent / tools / analysis /

statistics.py

Statistical significance tests.

Functions

Func

execute_statistical_test

Line 8
execute_statistical_test(**kwargs) -> Dict[str, Any]

Run statistical significance tests on experiment results.

Backs the tuiml_test_statistics tool. Supported tests: friedman, nemenyi, wilcoxon, paired_t, anova, friedman_aligned, quade. The pairwise tests (wilcoxon, paired_t) compare the first two algorithms in results.

Parameters

test
str
Name of the test to run (arrives via **kwargs, like all parameters below).
results
dict
{algorithm_name: [scores...]} mapping of per-fold scores.
significance_level
float = 0.05
Alpha level for significance.
higher_better
bool = True
Whether higher scores are better (pairwise tests only).

Returns

result
dict
On success: status ('success'), test and significant; most tests add statistic and p_value; pairwise tests add algorithms and a details dict of means/stds; nemenyi returns per-pair significance in details. On failure: status ('error'), error and optionally suggestion / error_type.