Introduction Version of Funnel Analysis Notebok #2070
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
View / edit / reply to this conversation on ReviewNB cetagostini commented on 2025-11-07T08:35:07Z Probably you don't want this, you are plotting the posterior of a transform variable predicted with another transform. If this was original scale or a simple example where we don't need to use transformations support the idea, but not here. |
|
View / edit / reply to this conversation on ReviewNB cetagostini commented on 2025-11-07T08:35:08Z The recover posterior is correct. juanitorduz commented on 2025-11-07T21:36:19Z :) |
|
View / edit / reply to this conversation on ReviewNB cetagostini commented on 2025-11-07T08:35:08Z What are you trying to do here? I feel estimate something is wrong here and looks like in sample fit? If so, I'll remove, we are working on a transform space this should say nothing.
ps: After this part, I'm not sure I follow, but I guess you are trying to show a recover of X4 (real variable not transform). Can be tricky, thats not in the original notebook, and never though about it. juanitorduz commented on 2025-11-07T21:37:18Z Here I am showing the x1 contribution and the posterior predictive mean to the likelihood |
|
View / edit / reply to this conversation on ReviewNB cetagostini commented on 2025-11-07T08:35:10Z You need to replace by:
# Create xarray with proper broadcasting for chain, draw, date dimensions
impressions_x1_values = X_train["impressions_x1"].values # shape: (date,)
saturation_beta_scaled = (
second_causal_mmm.idata.posterior.saturation_beta
* second_causal_mmm.scalers._target.item()
) # shape: (chain, draw)
# Broadcast to create (chain, draw, date) output
posterior_result = (
impressions_x1_values[None, None, :] * saturation_beta_scaled.values[:, :, None]
)
posterior_contribution_x1_over_x4 = xr.DataArray(
posterior_result,
dims=["chain", "draw", "date"],
coords={
"chain": second_causal_mmm.idata.posterior.chain,
"draw": second_causal_mmm.idata.posterior.draw,
"date": X_train.index,
},)
Sample can be complicated because transformations but you have the coefficient recover anyways. This should be equivalent. |
|
:) View entire conversation on ReviewNB |
|
Here I am showing the x1 contribution and the posterior predictive mean to the likelihood View entire conversation on ReviewNB |
The notebook https://www.pymc-marketing.io/en/latest/notebooks/mmm/mmm_upper_funnel_causal_approach.html is fantastic 🔥 ! We want to have a shorter version with more accessible comments and more detailed explanations for newcomers.
📚 Documentation preview 📚: https://pymc-marketing--2070.org.readthedocs.build/en/2070/