def add_adjusted_p_and_is_significant()

in spotify_confidence/analysis/frequentist/multiple_comparison.py [0:0]

• 67 lines of code
• 11 McCabe index (conditional complexity)

def add_adjusted_p_and_is_significant(df: DataFrame, **kwargs: Dict) -> DataFrame:
    n_comparisons = kwargs[NUMBER_OF_COMPARISONS]
    if kwargs[FINAL_EXPECTED_SAMPLE_SIZE] is not None:
        if kwargs[CORRECTION_METHOD] not in [
            BONFERRONI,
            BONFERRONI_ONLY_COUNT_TWOSIDED,
            BONFERRONI_DO_NOT_COUNT_NON_INFERIORITY,
            SPOT_1,
        ]:
            raise ValueError(
                f"{kwargs[CORRECTION_METHOD]} not supported for sequential tests. Use one of"
                f"{BONFERRONI}, {BONFERRONI_ONLY_COUNT_TWOSIDED}, "
                f"{BONFERRONI_DO_NOT_COUNT_NON_INFERIORITY}, {SPOT_1}"
            )

        groups_except_ordinal = [
            column
            for column in df.index.names
            if kwargs[ORDINAL_GROUP_COLUMN] is not None
            and column is not None
            and (column != kwargs[ORDINAL_GROUP_COLUMN] or kwargs[FINAL_EXPECTED_SAMPLE_SIZE] is None)
        ]
        df = groupbyApplyParallel(
            df.groupby(groups_except_ordinal + [kwargs[METHOD], "level_1", "level_2"], as_index=False, sort=False),
            lambda df: compute_sequential_adjusted_alpha(df, **kwargs),
        )
    elif kwargs[CORRECTION_METHOD] in [
        HOLM,
        HOMMEL,
        SIMES_HOCHBERG,
        SIDAK,
        HOLM_SIDAK,
        FDR_BH,
        FDR_BY,
        FDR_TSBH,
        FDR_TSBKY,
        SPOT_1_HOLM,
        SPOT_1_HOMMEL,
        SPOT_1_SIMES_HOCHBERG,
        SPOT_1_SIDAK,
        SPOT_1_HOLM_SIDAK,
        SPOT_1_FDR_BH,
        SPOT_1_FDR_BY,
        SPOT_1_FDR_TSBH,
        SPOT_1_FDR_TSBKY,
    ]:
        if kwargs[CORRECTION_METHOD].startswith("spot-"):
            correction_method = kwargs[CORRECTION_METHOD][7:]
        else:
            correction_method = kwargs[CORRECTION_METHOD]
        df[ADJUSTED_ALPHA] = df[ALPHA] / n_comparisons
        is_significant, adjusted_p, _, _ = multipletests(
            pvals=df[P_VALUE], alpha=1 - kwargs[INTERVAL_SIZE], method=correction_method
        )
        df[ADJUSTED_P] = adjusted_p
        df[IS_SIGNIFICANT] = is_significant
    elif kwargs[CORRECTION_METHOD] in [
        BONFERRONI,
        BONFERRONI_ONLY_COUNT_TWOSIDED,
        BONFERRONI_DO_NOT_COUNT_NON_INFERIORITY,
        SPOT_1,
    ]:
        df[ADJUSTED_ALPHA] = df[ALPHA] / n_comparisons
        df[ADJUSTED_P] = df[P_VALUE].map(lambda p: min(p * n_comparisons, 1))
        df[IS_SIGNIFICANT] = df[P_VALUE] < df[ADJUSTED_ALPHA]
    else:
        raise ValueError("Can't figure out which correction method to use :(")

    return df