Will Dean

icon indicating a website link https://williambdean.github.io/

icon company @pymc-labs icon location Data Science | Bayesian Statistics | Geospatial Analytics

Results 623 comments of Will Dean

Cannot save/load model

Can you load the nc file directly with arviz and share what are the attrs of InferenceData and the values of fit_data Dataset group

Cannot save/load model

I am unable to reproduce. Can you make a small reproducible example

Cannot save/load model

Hi @bravoila, can you inspect the mmm.idata.attrs of the two models and share the differences?

Cannot save/load model

> > [example.zip](https://github.com/user-attachments/files/18826277/example.zip) Please find attached @wd60622 > > [@williambdean](https://github.com/williambdean) were you able to reproduce? Please provide a minimal example that is not a zip file

Display comparison table more promimently

There is this. Let's consolidate: https://www.pymc-marketing.io/en/stable/guide/mmm/comparison.html

Geometric ad stock accepts alpha values outside its domain

Think this is a good idea. Probably want to use pymc / pytensor tooling though for the checks instead of numpy. The `pymc.logprob.utils.CheckParameterValue` can be used. Example: https://github.com/pymc-devs/pymc-experimental/blob/af91b423095eb37017eb84ded91c14f0eae08494/pymc_experimental/distributions/continuous.py#L331

Geometric ad stock accepts alpha values outside its domain

Thoughts @juanitorduz ?

MMM Case Study from PyData Global

Thanks for making that adjustment. Labels look better. Still have the comments on the notebook!

Add ROAS plot

Could the "channel_contributions" Deterministic be used for this? For instance, ```python import xarray as xr # (chain, draw, date, channel) channel_contributions = mmm.fit_result["channel_contribution"] # Divide with spends from the data...

Add ROAS plot

> Yes! We just need to be careful with the adstock contribution (similar as you did with the out-of-sample feature ;) ) Would this affect more than the first `l_max`...