Plot media contribution over time on a out of sample dataset

[NMaidanenko]

Hi all, is there any way to plot the media contribution plot over time on a out of sample dataset? Can anyone share their experience or suggest a solution ?

[wd60622]

Hi @NMaidanenko

Give this a try

import xarray as xr
import pandas as pd

from pymc_marketing.mmm import MMM


def new_data_media_contributions(X: pd.DataFrame, mmm: MMM) -> xr.DataArray:
    mmm._data_setter(X)
    with mmm.model:
        return pm.sample_posterior_predictive(
            mmm.idata,
            var_names=["channel_contributions"],
        ).posterior_predictive["channel_contributions"]

Be careful that the adstock effect from the last observations of the data set might want to be considered!

It is based off the sample_posterior_preditive method which does account the last observations of the fit data set:

pymc-marketing/pymc_marketing/mmm/delayed_saturated_mmm.py

Lines 1676 to 1688 in 05abba7

	if include_last_observations:
	X_pred = pd.concat(
	[self.X.iloc[-self.adstock.l_max :, :], X_pred], axis=0
	).sort_values(by=self.date_column)
	self._data_setter(X_pred)
	with self.model: # sample with new input data
	post_pred = pm.sample_posterior_predictive(
	self.idata, **sample_posterior_predictive_kwargs
	)
	if extend_idata:
	self.idata.extend(post_pred, join="right") # type: ignore

[wd60622]

On second thought, I think this can be done with the current implementation

mmm: MMM | DelayedSaturatedMMM = ...
var_names = ["channel_contributions"]
_ = mmm.sample_posterior_predictive(X_pred=..., var_names=var_names, extend_idata=True)

mmm.idata.posterior_predictive[var_names]

Give that a try and let me know if that works