plot_data#

SuperGLM.plot_data(
terms: Hashable | Sequence[Hashable] | None = None,
*,
kind: str = 'global',
ci: str | bool | None = 'pointwise',
X: object | None = None,
sample_weight: NDArray | None = None,
show_density: bool = True,
show_knots: bool = False,
show_bases: bool = False,
n_points: int = 200,
alpha: float = 0.05,
n_sim: int = 10000,
seed: int = 42,
centering: str = 'native',
) → dict[str, Any]#

Return plain data needed to recreate SuperGLM plots.

This is the data/export companion to plot(). It returns plain pandas DataFrames, NumPy arrays, and metadata dictionaries instead of a figure object, so users can rebuild charts in matplotlib, plotly, Excel, or another reporting system.

For main effects, the payload includes per-term fitted effects and, when requested, density overlays, spline knot positions, and basis contributions. For interactions, it includes the reconstructed effect data and, for continuous x continuous surfaces, optional density / HDR grid data when X and sample_weight are supplied.

With centering="native" (default), relativity values are the exponentiated fitted term contributions under the model’s identifiability constraint — not portfolio-average relativities. Pass centering="mean" for a reporting view where the geometric mean of relativities = 1.

A grouped OrderedCategorical is always exported on the expanded axis — one row per original level — where plot() defaults to the collapsed view and takes grouped_level_display=, which this method does not. That matters for a merged group, because the two payload pieces then live on different axes: effect["x_position"] holds the declared level values, while smooth_curve["x"] spans the axis the smooth was FITTED on, which for a merged group is its members’ mean position. A merge at either end of the ordering therefore leaves the exported curve short of the outermost exported markers, and a merged group’s two markers sit off the curve’s own band — correctly, since they are one parameter reported at two coordinates, but it reads like a coverage failure if you do not expect it. Join the two on the level table, not on x.

Examples

>>> payload = model.plot_data("DrivAge", X=X_train, sample_weight=w, show_knots=True)
>>> curve_df = payload["terms"][0]["effect"]
>>> knots_df = payload["terms"][0]["knots"]
On this page