Baseline discrepancy between train and predict

[raul-aily]

Hello everybody,
I am trying to obtain media impact on a test set. Obtaining something similar to the plot.create_media_baseline_contribution_df function. The problem I am seeing is that if I take the last 'baseline contribution' of the media_baseline_contribution_df and compare it with the 'baseline contribution' of the test, calculated as the prediction offset(The same way it is obtained in the response_curves), the values are considerably different. This problem is similar to the one exposed in issue #212 .