Plotting#

Term comparison across models. The per-model plotting methods (plot, plot_data, plot_diagnostics) live on SuperGLM and are documented on the model page.

superglm.plot_term_comparison

Public comparison entry point for labeled fitted-model overlays.

Appearance and observation support#

The default plots use blue effects, muted uncertainty bands, yellow support and orange markers for free levels. Matplotlib main-effect plots place support in a separate strip below each effect, with shared x limits and tick labels on the bottom axis. Unordered categories use points and error bars; ordered spline terms also show their fitted curve.

Pass X to show the observation distribution. With sample_weight, the strips show weighted density or weight per level. Without it, they show observation density or counts. Use show_density=False on model.plot() to hide them. With frequency weights, bar heights are replication totals. With prior weights, they are totals of the supplied precisions. The strips describe these supplied weights; they do not change the fitted model’s weight contract.

The returned Matplotlib figure remains editable, and plotting does not change global Matplotlib settings. Plotly uses the same palette; plotly_style can override its colours and sizes.

Main-effect figures keep constrained layout active so panels remain aligned when resized. Adjust its padding with fig.get_layout_engine().set(h_pad=0.1, w_pad=0.1); fig.subplots_adjust(...) is ignored while that engine is active. Manual ax.set_position(...) calls remove the chosen axis from automatic layout and survive subsequent draws.