Figure provenance and reproduction¶

Model diagrams from the white papers¶

shared-matrix-model.svg, sharing-across-runs.svg and gp-filtered-covariance.svg are copied without modification from shared-response-models/docs/whitepaper/figures/ at commit dd009e519919d38e978d54de9eedc1b1c8b14a9f. The source files were clean when copied. They contain schematic matrix patterns and model relationships, not research observations or fitted arrays.

Their source generator uses only Python's standard library. The accompanying R white paper and GP white paper retain their historical audit scope. The model mathematics page adapts their explanation to the current standalone package.

Kernel illustrations¶

kernel-families.svg, kernel-convolution.svg, kernel-parameters.svg, latent-timescale.svg and kernel-support.svg are generated by build_kernel_figures.py.

python -m pip install -e '.[plots]'
python scripts/build_kernel_figures.py

The response plots call the public kernel evaluators. Convolution illustrations use numerical integration on 3,601 response nodes. Matérn prior paths use a fixed seed of 42, 241 query times and a Cholesky factor with diagonal numerical jitter of \(10^{-10}\). They are synthetic teaching illustrations, not model fits or recovery demonstrations. SVG metadata omits the date and uses stable element IDs; exact rendering can still vary with plotting-library versions. Rebuilds also write PNG previews in the system temporary directory for visual inspection.

All figures include explanatory captions and image descriptions in their reader-facing pages. Mathematical notation is also presented as text/equations around the diagrams.