def prep_samples_for_cohort_grouping(
    *, df_samples, area_by, period_by, taxon_by="taxon"
):
    # Take a copy, as we will modify the dataframe.
    df_samples = df_samples.copy()

    # Fix "intermediate" or "unassigned" taxon values - we only want to build
    # cohorts with clean taxon calls, so we set other values to None.
    loc_intermediate_taxon = (
        df_samples[taxon_by].str.startswith("intermediate").fillna(False)
    )
    df_samples.loc[loc_intermediate_taxon, taxon_by] = None
    loc_unassigned_taxon = (
        df_samples[taxon_by].str.startswith("unassigned").fillna(False)
    )
    df_samples.loc[loc_unassigned_taxon, taxon_by] = None

    # Add period column.
    if period_by == "year":
        make_period = _make_sample_period_year
    elif period_by == "quarter":
        make_period = _make_sample_period_quarter
    elif period_by == "month":
        make_period = _make_sample_period_month
    else:  # pragma: no cover
        raise ValueError(
            f"Value for period_by parameter must be one of 'year', 'quarter', 'month'; found {period_by!r}."
        )
    sample_period = df_samples.apply(make_period, axis="columns")
    df_samples["period"] = sample_period

    # Add area column for consistent output.
    df_samples["area"] = df_samples[area_by]

    return df_samples

def _make_sample_period_month(row):
    year = row.year
    month = row.month
    if year > 0 and month > 0:
        return pd.Period(freq="M", year=year, month=month)
    else:
        return pd.NaT


def _make_sample_period_quarter(row):
    year = row.year
    month = row.month
    if year > 0 and month > 0:
        return pd.Period(freq="Q", year=year, month=month)
    else:
        return pd.NaT


def _make_sample_period_year(row):
    year = row.year
    if year > 0:
        return pd.Period(freq="Y", year=year)
    else:
        return pd.NaT