3daedc17ddab948752f3d6b90e56b177ea77ec42,nilearn/mass_univariate/permuted_least_squares.py,,permuted_ols,#Any#Any#Any#Any#Any#Any#Any#,252

Before Change

    h0_fmax = np.sort(np.ravel(np.concatenate(ret)))
    // convert scores into negative log10 p-values
    // TODO: to speed this up, we could threshold scores_original_data
    n_scores = n_descriptors * n_regressors
    ravelized_scores = np.ravel(scores_original_data)
    ret = joblib.Parallel(n_jobs=n_jobs)(joblib.delayed(_convert_to_pvalues)
          (h0_fmax, ravelized_scores[chunk])
          for chunk in gen_even_slices(
            n_scores + 1, min(n_scores, n_jobs)))
    pvals = np.concatenate(ret).reshape((n_regressors, n_descriptors))

    return pvals, scores_original_data, h0_fmax

After Change

    scores_as_ranks_parts, h0_fmax_parts = zip(*ret)
    h0_fmax = np.hstack((h0_fmax_parts))
    scores_as_ranks = np.zeros((n_regressors, n_descriptors))
    for scores_as_ranks_part in scores_as_ranks_parts:
        scores_as_ranks += scores_as_ranks_part
    // convert ranks into p-values
    pvals = (n_perm + 1 - scores_as_ranks) / float(1 + n_perm)

    return - np.log10(pvals), scores_original_data, h0_fmax[0]

Instances