diagnostics

Diagnostics namespace — posterior/prior predictive and residual plots.

This module exposes DiagnosticsPlots, which is the entry point for all time-series diagnostic plots on a fitted MMM model. It is accessed via mmm.plot.diagnostics (a DiagnosticsPlots instance backed by the model’s MMMIDataWrapper).

Examples

Construct the plotter directly from the model’s data wrapper:

from pymc_marketing.mmm.plotting.diagnostics import DiagnosticsPlots

dp = DiagnosticsPlots(mmm.data)

Posterior predictive — overlay posterior mean, HDI band, and observed target:

fig, axes = dp.posterior_predictive()

# Scaled units, narrower HDI, two columns per row
fig, axes = dp.posterior_predictive(original_scale=False, hdi_prob=0.8, col_wrap=2)

Prior predictive — same layout drawn from the prior:

fig, axes = dp.prior_predictive()

fig, axes = dp.prior_predictive(original_scale=False, hdi_prob=0.8, col_wrap=2)

Residuals over time — mean residual line with HDI band and a zero reference:

fig, axes = dp.residuals_over_time()

# Subset to one geo, custom styling
fig, axes = dp.residuals_over_time(
    hdi_prob=0.8,
    dims={"geo": "geo_a"},
    line_kwargs={"color": "red", "linestyle": "-."},
    hdi_kwargs={"color": "red"},
)

Residuals distribution — KDE of the posterior residual distribution:

fig, axes = dp.residuals_distribution(figsize=(8, 4))

# Custom quantile markers, collapse geo into one distribution
fig, axes = dp.residuals_distribution(
    figsize=(8, 4),
    quantiles=[0.1, 0.5, 0.9],
    aggregation="geo",
)

Posterior — 1-D marginal KDE for selected variables:

fig, axes = dp.posterior(["saturation_lam", "adstock_alpha"], figsize=(10, 8))

Prior vs posterior — overlaid prior and posterior KDEs:

fig, axes = dp.prior_vs_posterior(
    ["saturation_lam", "adstock_alpha"], figsize=(10, 8)
)

Classes

DiagnosticsPlots(data)

Time-series diagnostic plots for fitted MMM models.