Different prior distributions for different media channels on same coefficient

[l-d96]

trafficstars

Hi,

thank you for your hard work!

Reading through the documentation and the code base I can see how we can set custom priors for the coefficients of the regression.

However for a single coefficient, let's say lag_weight, I can set the distribution to be numpyro.distributions.Beta. If I have several media channels, they will all have a Beta distributions, albeit with different parameters depending on what I pass in the custom_priors dictionary.

Is there any way to add different types of distributions as priors for a single coefficients? E.g. with 3 channels: TV, Paid Social, Paid Search, set the lag_weight to be Beta distribution for TV, HalfNormal for Paid Social and Gamma for Paid Search?

Again, thanks a lot for this awesome project!

[AllardJM]

@l-d96 Did you ever find a way to do this? Even at the basic level, how do you pass in different parameters for the same distribution for each of the media channels?

[l-d96]

@AllardJM unfortunately no. For passing different parameters for each media channels, you can just pass an array containing the paramaters for each argument of the distribution.

You can check other way to pass different parameters here: Custom priors in LightweightMMM