Docs: Add visual documentation for adstock mode parameter (Before/After/Overlap)

[daimon-pymclabs]

Issue generated by @drbenvincent from an EAP advisory call discussion.

Problem

The adstock transformations (DelayedAdstock, WeibullPDFAdstock, etc.) support a mode parameter with three convolution modes:

• After — traditional trailing decay (standard adstock)
• Before — "excitement" / ramp-up effect (forward-looking)
• Overlap — pull-forward and pull-backward

This parameter is defined in the code and mentioned in docstrings, but there are no plots demonstrating the visual effect of changing mode. Users currently see plots for different alpha/theta/k/lambda values but nothing showing what mode="Before" or mode="Overlap" actually looks like.

This was highlighted during a technical discussion where a production MMM team was unaware this feature existed. The mode parameter is particularly useful for modeling holiday/event effects — e.g., a ramp-up in purchasing before a holiday followed by decay after — but is effectively undiscoverable without visual documentation.

Suggested approach

Either (or both):

1. Docstring plots — Add example plots to the adstock class docstrings showing the same transformation with mode="After" vs mode="Before" vs mode="Overlap" side by side
2. Dedicated docs page or section — Extend the existing adstock functions guide notebook (docs/source/notebooks/mmm/adstock_functions_guide.ipynb) with a section on the mode parameter, including:
   • Visual comparison of all three modes
   • Practical use cases (e.g., holiday ramp-up, event anticipation effects)
   • How to apply this to event/holiday indicator variables

Context

The delayed adstock page already shows plots for varying theta and k/lambda, but the mode parameter is absent from all visualizations. The Weibull PDF adstock page similarly lacks mode demonstrations.

🤖 Generated with Claude Code

[williambdean]

There is something here: https://www.pymc-marketing.io/en/stable/api/generated/pymc_marketing.mmm.transformers.batched_convolution.html#pymc_marketing.mmm.transformers.batched_convolution

[daimon-pymclabs]

Good catch — thanks for the pointer. The batched_convolution API docs do show the mode parameter exists. The gap this issue targets is more about visual discoverability in the higher-level adstock class docs and guide notebooks: the existing adstock functions guide shows plots for varying alpha/theta/k/lambda but has no plots demonstrating what mode="Before" or mode="Overlap" actually looks like on the transformed curve. The ask is to add side-by-side mode comparison plots there (and/or in the class docstrings) so users exploring adstock don't have to find their way to the lower-level batched_convolution page to learn about it.