How to use the pastas.utils.get_sample function in pastas

series = series.ffill(limit=limit)
    elif fill_method == 'bfill':
        series = series.bfill(limit=limit)
    elif fill_method == 'nearest':
        if limit == 0:
            # limit=0 is a trick to only use each measurements once
            # only keep days with measurements
            series = series.dropna()
            # generate an index at the 14th and 28th of every month
            buf = pd.to_timedelta(8, 'd')
            ref_index = pd.date_range(series.index.min() - buf,
                                      series.index.max() + buf)
            mask = [(x.day == 14) or (x.day == 28) for x in ref_index]
            ref_index = ref_index[mask]
            # only keep the days that are closest to series.index
            ref_index = get_sample(ref_index, series.index)
            # and set the index of series to this index
            # (and remove rows in series that are not in ref_index)
            series = series.reindex(ref_index, method=fill_method)
            select14or28 = False
        else:
            # with a large limit (larger than 6) it is possible that one measurement is used more than once
            series = series.dropna().reindex(series.index, method=fill_method,
                                             limit=limit)
    else:
        series = series.interpolate(method=fill_method, limit=limit,
                                    limit_direction='both')

    # and select the 14th and 28th of each month (if needed still)
    if select14or28:
        mask = [(x.day == 14) or (x.day == 28) for x in series.index]
        series = series.loc[mask]

series = series.ffill(limit=limit)
    elif fill_method == 'bfill':
        series = series.bfill(limit=limit)
    elif fill_method == 'nearest':
        if limit == 0:
            # limit=0 is a trick to only use each measurements once
            # only keep days with measurements
            series = series.dropna()
            # generate an index at the 14th and 28th of every month
            buf = to_timedelta(8, 'd')
            ref_index = date_range(series.index.min() - buf,
                                   series.index.max() + buf)
            mask = [(x.day == 14) or (x.day == 28) for x in ref_index]
            ref_index = ref_index[mask]
            # only keep the days that are closest to series.index
            ref_index = get_sample(ref_index, series.index)
            # and set the index of series to this index
            # (and remove rows in series that are not in ref_index)
            series = series.reindex(ref_index, method=fill_method)
            select14or28 = False
        else:
            # with a large limit (larger than 6) it is possible that one measurement is used more than once
            series = series.dropna().reindex(series.index, method=fill_method,
                                             limit=limit)
    else:
        series = series.interpolate(method=fill_method, limit=limit,
                                    limit_direction='both')

    # and select the 14th and 28th of each month (if needed still)
    if select14or28:
        mask = [(x.day == 14) or (x.day == 28) for x in series.index]
        series = series.loc[mask]

update_observations = False
        for key, setting in zip([tmin, tmax, freq], ["tmin", "tmax", "freq"]):
            if key != self.settings[setting]:
                update_observations = True

        if self.oseries_calib is None or update_observations:
            oseries_calib = self.oseries.series.loc[tmin:tmax]

            # sample measurements, so that frequency is not higher than model
            # keep the original timestamps, as they will be used during
            # interpolation of the simulation
            sim_index = self.get_sim_index(tmin, tmax, freq,
                                           self.settings["warmup"])
            if not oseries_calib.empty:
                index = get_sample(oseries_calib.index, sim_index)
                oseries_calib = oseries_calib.loc[index]

            if not update_observations:
                # tmin, tmax and freq are equal to the settings
                # so we can set self.oseries_calib to improve speed of next run
                self.oseries_calib = oseries_calib
        else:
            oseries_calib = self.oseries_calib
        return oseries_calib