generating adstock with 'pre model time period' data?

[lorenze3]

Howdy, I'm encountering a steady and regular decrease in base volume to zero.

I can reduce media contribution (by cutting the average of the priors) quite a bit, but I'm wondering if having non-zero adstock in the initial modeling weeks might help prevent this (as the model 'as is' starts with no build up of adstock in the initial weeks, the media drivers all tend to increase through the initial modelling period, which might be artificially depressing the base driver contribution in the later time periods of the model).

When I've worked with adstock and saturation as a hyper parameter tuning problem, it is straigthforward to compute adstock for the first quarter or two of media data and then only estimate the model after that pre-period. But I'm not sure if it's possible in this framework to do this same?

I'd appreciate the feature, of course, but also any advice on where to modify the existing codebase to try to get this to work.

Thanks for the work on this so far!