Hi!

I am trying to incorporate media "burn-in" effects by having n_media_times = n_times + max_lag, as described in the API reference (media attribute). I suspect that there is an issue in the code though as the DataFrameInputDataBuilder.with_media method sets time_col. In contrast, the DataFrameInputDataBuilder.with_organic_media sets media_time_col.

Thus when I have my input dataframe that has max_lag more rows in the beginning for my media data than for my kpi I get the following error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[107], line 13
      1 data_loader = DataFrameDataLoader(
      2     df=df,
      3     coord_to_columns=coord_to_columns,
   (...)
     11     #organic_frequency_to_channel=organic_frequency_to_channel,
     12 )
---> 13 data = data_loader.load()

File /opt/conda/lib/python3.10/site-packages/meridian/data/load.py:759, in DataFrameDataLoader.load(self)
    754   # So now we can use one of the channel mapper dicts to get the canonical
    755   # channel names for each column.
    756   media_channel_names = [
    757       self.media_to_channel[c] for c in self.coord_to_columns.media
    758   ]
--> 759   builder.with_media(
    760       self.df,
    761       media_execution_columns,
    762       media_spend_columns,
    763       media_channel_names,
    764       self.coord_to_columns.time,
    765       self.coord_to_columns.geo,
    766   )
    768 if (
    769     self.reach_to_channel is not None
    770     and self.frequency_to_channel is not None
   (...)
    774   # listed in `reach`, `frequency`, and `rf_spend` are already validated
    775   # to correspond to the same channels, in user-given order.
    776   reach_columns = list(self.coord_to_columns.reach)

File /opt/conda/lib/python3.10/site-packages/meridian/data/data_frame_input_data_builder.py:381, in DataFrameInputDataBuilder.with_media(self, df, media_cols, media_spend_cols, media_channels, time_col, geo_col)
    379 media_spend_data = media_df.set_index([geo_col, time_col])[media_spend_cols]
    380 media_spend_data.columns = media_channels
--> 381 self.media_spend = (
    382     media_spend_data.stack()
    383     .rename(constants.MEDIA_SPEND)
    384     .rename_axis([
    385         constants.GEO,
    386         constants.TIME,
    387         constants.MEDIA_CHANNEL,
    388     ])
    389     .to_xarray()
    390 )
    391 return self

File /opt/conda/lib/python3.10/site-packages/meridian/data/input_data_builder.py:338, in InputDataBuilder.media_spend(self, media_spend)
    336 self._media_spend = self._normalize_coords(media_spend, constants.TIME)
    337 self.geos = self.media_spend.coords[constants.GEO].values.tolist()
--> 338 self.time_coords = self.media_spend.coords[constants.TIME].values.tolist()

File /opt/conda/lib/python3.10/site-packages/meridian/data/input_data_builder.py:85, in InputDataBuilder.time_coords(self, value)
     83   raise ValueError('`times` coords must be unique.')
     84 if self.time_coords is not None and set(self.time_coords) != set(value):
---> 85   raise ValueError(f'`times` coords already set to {self.time_coords}.')
     86 if self.media_time_coords is not None and not set(value).issubset(
     87     self.media_time_coords
     88 ):
     89   raise ValueError(
     90       '`times` coords must be subset of previously set `media_times`'
     91       ' coords.'
     92   )

ValueError: `times` coords already set to ['2022-07-21', '2022-07-22', ...

It seems that there is a mismatch in the time value, which stems from the fact that my media data has more rows than my remaining data.