search

sibylhe / mmm_stan

Python/STAN Implementation of Multiplicative Marketing Mix Model, with deep dive into Adstock (carry-over effect), ROAS, and mROAS
MIT License
352 stars 162 forks source link

Different results between Jupyter Notebook and Readme #50

Closed Steven-Livingstone closed 3 years ago

Steven-Livingstone commented 3 years ago

First of all AWESOME STUFF!

However, in an effort to reproduce your results locally I spotted a few differences between your Jupyter notebook results and the ones given in the readme.

Value stan_mmm.ipynb readme.md
MAPE 26.06330645337451 8.63
RMSE (log-log model) 0.13647800963534104 0.04977
MAPE (multiplicative model) 20.313393234793644 15.71

Have the Readme results been generated using different priors/parameters, or am I missing something obvious?

sibylhe commented 3 years ago

Don't bother, it doesn't really matter. They were generated from different datasets. The dataset published is tweaked, and the Jupyter notebook is done on top of it.

Steven-Livingstone commented 3 years ago

Ah, that makes sense then.

Thanks for the quick clarification. :)