search

google / lightweight_mmm

LightweightMMM 🦇 is a lightweight Bayesian Marketing Mix Modeling (MMM) library that allows users to easily train MMMs and obtain channel attribution information.
https://lightweight-mmm.readthedocs.io/en/latest/index.html
Apache License 2.0
829 stars 172 forks source link

media_priors #276

Open rajat-barve opened 7 months ago

rajat-barve commented 7 months ago

Hi All,

I have some questions about setting the media_priors.

1) Is there a way to change the default HalfNormal distribution to something else, say, Gamma? I ask because suppose I have reason to believe that the PDF should peak at certain positive value (but we know that the HalfNormal is zero-biased).

2) Why do we 1-center the costs? What happens if I do not 1-center it? Especially when the costs do not even go into the model as training data but are only used to set the default media_priors. I can understand the reason behind 1-centering the dependent and independent variables of the model. But why the media_priors?

jtitusj commented 7 months ago

I have the same questions here.

For question 2, I'm not sure what the underlying principle for 1-centering. In our case, 1-centering results in a higher marketing contribution as compared to dividing the media spend by the max value.