DiagnosticsPlots.residuals_distribution#

DiagnosticsPlots.residuals_distribution(quantiles=None, aggregation=None, idata=None, dims=None, figsize=None, backend=None, return_as_pc=False, dist_kwargs=None, **pc_kwargs)[source]#

Plot the posterior distribution of residuals using arviz-plots.

Uses azp.plot_dist (KDE) with quantile reference lines via azp.add_lines. The distribution is computed over ["chain", "draw", "date"] plus any dimensions in aggregation, so extra model dims (e.g. "geo") are structural facet dims by default.

Parameters:

quantileslist[float], optional: Quantile probabilities to mark as vertical reference lines. Default [0.025, 0.5, 0.975]. Each value must be in [0, 1].
aggregationlist[str] or str, optional: Extra custom dimension names to collapse into the distribution (added to sample_dims beyond ["chain", "draw", "date"]). A single string is accepted and treated as [aggregation]. Example: aggregation="geo" or aggregation=["geo"] merges geo panels into one combined distribution. Default None — extra dims are structural facet dims.
idataaz.InferenceData, optional: Override instance data.
dimsdict[str, Any], optional: Subset dimensions applied before plotting.
figsizetuple[float, float], optional: Figure size forwarded via figure_kwargs.
backendstr, optional: Rendering backend (e.g. "matplotlib"). Non-matplotlib backends require return_as_pc=True.
return_as_pcbool, default False: Return the raw PlotCollection instead of (Figure, NDArray[Axes]).
dist_kwargsdict, optional: Extra kwargs forwarded to azp.plot_dist.
**pc_kwargs: Forwarded to azp.plot_dist (e.g. figure_kwargs).

Returns:

PlotCollection or tuple[Figure, NDArray[Axes]]

Examples

fig, axes = mmm.plot.diagnostics.residuals_distribution()
fig, axes = mmm.plot.diagnostics.residuals_distribution(
    quantiles=[0.05, 0.5, 0.95], aggregation=["geo"]
)