find_optimal_budgets geo level about lightweight_mmm HOT 8 CLOSED

At the moment optimisation is provided at channel level always, this is mainly due that bounds can be very tricky to use at geo level and the actual shifts of channel budget between geos at an operational level sometimes is not desired by the user.

We assume the average ratios between geos seen in the historical data and apply optimisation at channel level using such ratio.

It is something we can add for sure, but it is also not in our most immediate pipeline (unless there is a lot of demand for it).

Will keep this one open though as we should be able to add it at some point.

from lightweight_mmm.

1. so, can I use the preoptimization budget ratios of goe's and apply to post opt budget and get geo level post opt budget?
2. And won't the most effective channel geo budget ratio increase while lease effect channel geo decreases?

from lightweight_mmm.

1. Yes! Feel free to check how the ratios are calculated in the code to replicate.
2. Not sure what you mean by this one. Could you explain further?

from lightweight_mmm.

Hi @pabloduque0,

I couldn't find the ratios logic, can you please help me ?

from lightweight_mmm.

Ratios can be calculated with something like the folloiwng:

average_per_time = media_mix_model.media.mean(axis=0)
geo_ratio = average_per_time / jnp.expand_dims(average_per_time.sum(axis=-1), axis=-1)

from lightweight_mmm.

Hi @pabloduque0,
I saw your great discussion, but I have some questions regarding this.
First, is the calculating logic you introduced above just for getting ratios at pre-optimization?
Moreover, it seems that the function find_optimal_budgets in the current version cannot consider the custom budget ratio of each geo at pre-optimization. Is my understanding correct?

Thank you.

from lightweight_mmm.

Hello @rio-0217 !
Yes, the logic is for calculating historical average ratio between geos.
That is correct! find_optimal budgets currently does not support custom "geo ratios" although that is something we could add.

from lightweight_mmm.

Hello @pabloduque0,
Understood. I appreciate your kind reply.
Then I wonder if we can just add geo_ratio in the find_optimal_budgets function as a new argument, we can conduct optimization with custom geo_ratios. Is that make sense?

from lightweight_mmm.