pymc-marketing-mmm

name: pymc-marketing-mmm description: Implement Marketing Mix Models (MMM) using the pymc-marketing Bayesian library. Covers model setup, adstock/saturation transformations, fitting, diagnostics, budget allocation, lift test calibration, causal inference, and model evaluation. Use when building or improving MMM models with pymc-marketing, navigating pymc-marketing documentation, choosing the right MMM notebook or API reference, or answering any pymc-marketing MMM implementation question. Combine with the pymc-modeling skill for advanced Bayesian modeling questions.

MMM with PyMC-Marketing

Implement Marketing Mix Models (MMM) end-to-end using the pymc-marketing Bayesian library.

When to Use This Skill

Load this skill when:

• User is implementing, debugging, or extending an MMM with pymc-marketing
• User asks about adstock, saturation, channel contributions, ROAS, or budget allocation in a pymc-marketing context
• User needs to navigate pymc-marketing documentation to find the right example or API reference
• User is calibrating an MMM with lift tests or applying causal inference techniques
• Any pymc-marketing MMM question arises

If available, combine with pymc-modeling skill for advanced Bayesian modeling, prior selection, or MCMC diagnostics questions.

Step 1: Clarify the Task

Identify which MMM sub-task the user needs before diving in:

Step 2: Navigate Documentation

Use these canonical references. Always fetch the relevant page to provide up-to-date code and API usage — pymc-marketing evolves rapidly (currently v0.19.3).

Primary Entry Points

Task-to-Notebook Map

Full notebook reference: see notebook-map.md

Base URL for all notebooks: https://www.pymc-marketing.io/en/stable/notebooks/mmm/

Quick lookup for most common tasks:

Full URL pattern: https://www.pymc-marketing.io/en/stable/notebooks/mmm/{slug}.html

Step 3: Core MMM Implementation

Installation

pip install pymc-marketing

Minimal Working Model

import pandas as pd
from pymc_marketing.mmm import (
    GeometricAdstock,
    LogisticSaturation,
    MMM,
)

mmm = MMM(
    adstock=GeometricAdstock(l_max=8),
    saturation=LogisticSaturation(),
    date_column="date_week",
    channel_columns=["x1", "x2"],
    control_columns=["event_1", "event_2", "t"],
    yearly_seasonality=2,
)

Key Model Components

Fit the Model

idata = mmm.fit(X, y)

Standard Post-Fitting Workflow

1. Prior predictive check → mmm.sample_prior_predictive(X) — validate priors before seeing data
2. MCMC diagnostics → inspect r_hat and ess_bulk in idata.sample_stats
3. Posterior predictive → mmm.sample_posterior_predictive(idata)
4. Channel contributions → mmm.plot_channel_contribution_share_hdi()
5. ROAS summary → mmm.get_channel_contributions_share_summary()
6. Budget optimization → mmm.optimize_budget(budget, ...)

For MCMC diagnostics, prior elicitation, and convergence details, load the pymc-modeling skill.

Step 4: Fetch Live Documentation

When implementing any specific feature, fetch the corresponding notebook for current, runnable code:

https://www.pymc-marketing.io/en/stable/notebooks/mmm/{notebook-slug}.html

Do not rely solely on training knowledge — the API changes frequently across versions.

References

• Full Notebook Map — Complete task-to-notebook lookup table
• Getting Started
• Example Gallery
• API Reference
• GitHub Repository