What should I use for media_prior argument in mmm.fit function?

[AruunnAnanda]

Hi there,

I understand that for the media prior argument I need to pass one value for each media spend channel in the form of an array.

Now my question is what should that value be?

Currently, I used a randomly typed number for all the channels as you can see below and executed the code. Is this wrong? What is your recommendation?

Thanks!

[michevan]

Hi AruunnAnanda

If you have data for the total cost of the media on each channel, we would recommend using those values here for your media priors, after normalizing them with a CustomScaler as shown in the example Colabs!

[danivapp]

How is the half-normal distribution generated form only the grand total? Is there any documentation regarding this?

[greenfrog555]

This is the document regarding how half-normal distribution with cost adjusted scale works. The equation is under the "media effect" section. Let us know whether it is clear.

[danivapp]

Thanks @greenfrog555, i have seen this durin my research. Half-normal is parametrized by only one parameter - its scale σ, since the location is fixed to μ=0 by the very definition of this distribution. Is the total cost passed as the scale parameter then? I seem to be missing something trivial here.

[greenfrog555]

Yes. The scaled cost for each channel will be passed as the scale in the Half-Normal distribution. The total cost across all the channels will be scaled with this function in our sample Colab. "cost_scaler = preprocessing.CustomScaler(divide_operation=jnp.mean, multiply_by=0.15)" The total cost is scaled by the mean and times a multiplier whose default is 0.15 but you can adjust it.

[aruizartg]

Hello,

I entered this discussion because I have a similar question regarding the media_prior parameter. Should we always calculate the total costs based on the spend of each media channel? For instance, if we're using different metrics for the media parameter (such as impressions, GRPs for TV, etc.), do we still determine the media_prior by summing up the spends, or should we aggregate the metrics used to define the media parameter in this scenario?

I realize this might be a straightforward question, but I want to be absolutely certain. Thank you in advance! 😄