How-to Guides

These guides are practical, task-focused instructions for AMMM V2. They assume you are running the MMMBaseDriverV2 workflow and reviewing outputs in the stage-based results structure (00_run_metadata/ through 80_interpretation/).

Recommended reading order:

1. Configure and run a baseline model.
2. Review diagnostics and gating outcomes.
3. Move to optimisation, scenarios, and advanced modelling choices.

• Configuration: Set up config.yml, sampling settings, Prophet options, and gating policy.
• Running The Model: Execute runme.py or the V2 driver API and verify stage outputs.
• Interpreting Results: Read fit, decomposition, diagnostics, and optimisation artefacts.
• Optimisation: Run single-period budget optimisation and interpret outputs.
• Multi-Period Optimisation: Plan budgets across multiple periods with seasonality.
• Event Buildup Modelling: Model pre-event demand shifts (for example BFCM buildup).
• Hypothesis Testing: Test directional hypotheses on control effects using posterior outputs.
• Cross-Validation: Evaluate predictive performance using LOO and hold-out checks.
• Prior Calibration: Build prior suggestions from historical AMMM runs.
• Lift Test Calibration: Incorporate experimental lift evidence into MMM parameter learning.
• Cache Sessions: Manage PyTensor cache/session behaviour across repeated runs.
• V2 Migration: Migrate legacy workflows to the V2 architecture.
• Docstring Style: Internal guide for consistent Google-style docstrings.