You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While digging deeper into the specifics, I was wondering how the decomposition distance is calculated, what is the rationale behind it and how you decide which pareto-optimal solution to pick when weighing MAPE against decomposition distance?
Thanks in advance!
Tim
The text was updated successfully, but these errors were encountered:
Hey, thanks for trying it out! For the decomposition distance, please see here. Regarding the model selection, Pareto optimality gives you the best options trading off model fit (NRMSE) and business fit (DECOMP.RSSD). But you're still getting a set of results and need to decide which one to pick. It's our decision to leave users these choices so that is not a blackbox solution. Selection guide could be:
Experimental results (FB lift, Geo test etc)
previous MMM results that are considered good fit
any attribution results that are considered good fit.
In the end, we believe experiments and calibrating the model using experimental results is the way to go. Hope it helps
Hi Facebook colleagues,
amazing package!
While digging deeper into the specifics, I was wondering how the decomposition distance is calculated, what is the rationale behind it and how you decide which pareto-optimal solution to pick when weighing MAPE against decomposition distance?
Thanks in advance!
Tim
The text was updated successfully, but these errors were encountered: