Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Decomposition Distance #135

Closed
TimKreienkamp opened this issue Aug 12, 2021 · 2 comments
Closed

Decomposition Distance #135

TimKreienkamp opened this issue Aug 12, 2021 · 2 comments

Comments

@TimKreienkamp
Copy link

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

@gufengzhou
Copy link
Contributor

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

@gufengzhou
Copy link
Contributor

please re-open if necessary

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants