Recovering adstock parameters from simulated dataset?

[NataliaVConnolly]

Hi All,

I am trying to reconcile the output from lightweight MMM with known inputs from a simulated dataset. I used https://github.com/facebookexperimental/siMMMulator/blob/main/siMMMulator-website/docs/step_by_step_guide.md to create a simple dataset with 3 media variables and known adstock lambdas of (0.1, 0.2, 0.3). I also applied alpha saturation = (2,2,2), and gamma saturation (0.1, 0.2, 0.3).

When I used lightweight MMM on this dataset, I assumed mmm.trace['lag_weight'] would recover my lambdas. Instead, I am getting these (relatively) very large values:

	channels	       adstock_decay
	clicks_Search	        0.702757
	impressions_TV	        0.717533
	impressions_Facebook	0.719859

Could someone please help me understand why this could be the case, and what I could do to get something closer to the true adstock values?

[8-u8]

@NataliaVConnolly
Hi, I want to reproduce your result. Can you share your LightweightMMM code, notebook or dummy data created by siMMMulator?