clarification about model input parameters about lightweight_mmm HOT 4 CLOSED

Thanks a lot @pabloduque0 it's much clearer now :)

Cheers

from lightweight_mmm.

Hello @RiccardoBarb !

1. Media data can contain impressions, clicks or spend (cost). But if is indeed impressions or clicks it wont contain cost as you said.
2. A single vector indicating the total cost of the channel in the data.
3. When we talk about costs we refer total costs. For optimisation and other parts we refer to price as the price per unit of an impression/click.

Hope it helps! Let me know if anything needs further clarification!

from lightweight_mmm.

hi @pabloduque0 thanks for your response!

What would be the recommended variable to use for media_data? would it be better to use spend or clicks/impressions?.
Also if I decide to use spend as media_data, then wouldn't cost be redundant information? would it still be needed to fit the model?
As far as the second answer is concerned, what if I have multiple media channels? Should I insert a single vector for each channel (which, in this case, is in fact equivalent to a matrix)?

Thanks

from lightweight_mmm.

Between clicks and impressions, impressions are preferred.

In reality the costs are used for the media prior (if you are referring to mmm.fit(..., media_prior=costs) therefore costs are still relevant as they serve as the prior. If you want to use a different prior then you mostly wont need costs. The only place where costs are "mandatory" is for ROI calculation (for obvious reasons).

Apologies if that was not clear before, what I mean is a single vector indicating the total cost each channel in the data, with ONE value per channel, so the total costs. So always one vector with one value per media channel. Eg. for three channels your costs could be jnp.array([1, 2, 3]).

from lightweight_mmm.