scikit-posthocs

0.12.0 verified Mon Apr 27 auth: no python

Statistical post-hoc analysis and outlier detection algorithms for Python. Provides pairwise multiple comparison tests (e.g., Nemenyi, Conover, Dunn, Bonferroni) and outlier detection (e.g., IQR, Grubbs). Current version: 0.12.0. Release cadence is irregular, major updates every 1-2 years.

pip install scikit-posthocs

Common errors

error ImportError: cannot import name 'posthoc_nemenyi_friedman' from 'scikit_posthocs' ↓

cause Misspelled package name or wrong import path. The package is 'scikit-posthocs' on PyPI but the module is 'scikit_posthocs' (underscore, not hyphen).

fix

Install with pip install scikit-posthocs and import from scikit_posthocs (underscore).

error ValueError: The number of groups must be > 2 ↓

cause Post-hoc tests require at least 3 groups (treatments). You passed data with only 2 columns.

fix

Ensure your data array has at least 3 columns (e.g., data.shape[1] >= 3).

error TypeError: posthoc_nemenyi_friedman() missing 1 required positional argument: 'data' ↓

cause In old code (pre-v0.8.0), the first argument was 'a'. Now it's 'data'.

fix

Use posthoc_nemenyi_friedman(data=your_data) or just your_data as first argument if using version >=0.8.0.

Warnings

breaking In v0.8.0+, the function signature changed: argument names switched from 'a' to 'data' and 'group' to 'group_col' (where applicable). Old code using positional arguments may break. ↓

fix Update to use named arguments: posthoc_nemenyi_friedman(data, group_col='groups')

breaking The function 'outliers_iqr' was removed in v0.10.0. Use 'outlier_test' instead or handle IQR manually. ↓

fix Replace outliers_iqr(data) with outlier_test(data, method='iqr')

deprecated The 'sign_array' function is deprecated since v0.11.0. Use 'sign_plot' instead for visualization. ↓

fix Replace sign_array(...) with sign_plot(...) (same arguments).

gotcha Input data must be a 2D array-like (n_subjects x n_treatments). If you pass a 1D array or accidentally transpose, results will be wrong without error. ↓

fix Always verify data shape: data.shape should be (subjects, treatments).

Imports

posthoc_nemenyi_friedman

wrong

from scikit_posthocs.posthocs import posthoc_nemenyi_friedman

correct

from scikit_posthocs import posthoc_nemenyi_friedman

The submodule posthocs does not exist; all functions are in top-level.

Quickstart

Perform Nemenyi post-hoc test after Friedman test. Input data: rows = subjects, columns = treatments. Output: pairwise p-value matrix.

import numpy as np
from scikit_posthocs import posthoc_nemenyi_friedman

data = np.array([   [1,2,3],
                     [2,3,4],
                     [3,4,5],
                     [4,5,6] ])

result = posthoc_nemenyi_friedman(data)
print(result)