Robyn model summary

[anuragvimal]

Hi Team,

I am working on 40 media variables for an MMM exercise. I want to have a look at feature importance and model summary as an output with a corresponding p-value. This can help me in discarding features that are insignificant. Is this possible in Robyn?

If not, then generally what are the best principles adopted for feature selection in Robyn? Thanks in advance.

[anuragvimal]

@kyletgoldberg Can you please help? Thanks.

[kyletgoldberg]

Hi - see this discussion about p-values & regularized regression here for why we do not have a p-value exposed. #131

In terms of feature selection, first and foremost we typically recommend to have no more than 1 independent variable for every 7-10 observations in your dataset. So if you are using daily data you can have quite a few vars. In terms of narrowing down your features, take a look at how impactful the variables are and also think about which vars are least impactful, and either may need to be grouped with another, or don't really explain much anyway.