Reproduce the visual showcases • grip

The construction traces show GRIP introducing vertices and refining their positions. The saddle animations rotate fixed saved coordinates. Use the controls below to play or stop motion; every display starts with a still view. Stopping returns to that preview. On GitHub, follow the README’s showcase link to reach this page. Without JavaScript, the still views and recipes remain usable.

These larger showcases are separate from the small, executable Getting Started and synthetic graph examples.

Triangle construction

Finished level-6 Sierpinski triangle layout with nested triangular holes.

A level-6 Sierpinski triangle, recorded with trace.grip() in two dimensions, preset = "carpet", and seed 1. This is a construction sequence, not elapsed physical time. Frames are aligned to the generator’s reference coordinates and displayed in a common frame so changes in orientation and overall scale do not obscure the refinement.

The maintained asset generator contains the graph construction, complete solver call, frame selection, alignment, symmetry convention, limits, colors, and export settings. The output is 600 by 600 pixels, 4 frames per second, with a hold on the final layout. Its poster is taken from that same final frame.

From a checkout of the version whose display you want to reproduce:

git clone https://github.com/pgajer/grip.git
cd grip
# For a historical display, check out its recorded commit before rendering.
Rscript -e 'install.packages(c("pkgload", "magick", "png", "rmarkdown", "knitr"))'
make readme-assets

The build record records the source commit, package version, seeds, settings, and R session for these assets. A checkout also needs a C++17 toolchain. The magick R package may need ImageMagick development libraries when installed from source. Rendering is an explicit task; installing grip does not run these large traces.

For a smaller example using the public generator:

library(grip)
edges <- edges.sierpinski.triangle(3)
trace <- trace.grip(edges, n = max(edges), dim = 2,
                    preset = "carpet", trace = "level", seed = 1)
plot.layout(trace$final, edges = edges, pch = 16)

This smaller level-3 graph does not reproduce the level-6 asset. For tracing and scoring your own run, follow Tracing and Diagnosing Layouts.

Carpet construction

Completed level-4 Sierpinski carpet layout with its recursive square holes.

The same make readme-assets command builds this level-4 carpet using preset = "carpet", seed 24, and a round trace in two dimensions. The generator aligns frames to the recursive grid and uses shared plotting limits. The synthetic graph guide provides smaller executable examples for inspecting holes, graph lengths, and reference coordinates.

Saved saddle rotations

Three saved saddle configurations viewed with a common triangulation and coordinate axes.

These are frozen display configurations of 1,000 corresponding vertices and 1,977 triangles. Their saved labels are “Original saddle”, “Metric MDS”, and “MDS + edge-KK”. The fixture does not record the solver version, objective, initialization, convergence, or alignment transform used to produce the fitted coordinates. In particular, the historical “Metric MDS” label does not establish that this display used the current stress-minimizing metric.mds(). Use it to inspect the saved shapes, not to compare current solver accuracy.

Overlay of the saved fitted configurations in blue and orange with the generating saddle in gray.

Both animations rotate fixed, already aligned coordinates around the vertical axis, at 20 frames per second and 12 seconds per revolution. They share their camera convention, scales, and triangulation; the overlay uses blue and orange for the saved fits and gray for the generating saddle. Motion does not show optimization or provide an accuracy measure.

The fixture manifest records what is known and what is missing. To replay the stored coordinates without rerunning fits:

make readme-saddle-animation
make saddle-overlay-animation

These explicit WebGL rendering targets need ivue with layer3D.axes(), camera.zup(), and animate.frames(), plus rgl, htmlwidgets, Pandoc, Node.js, Playwright, and a compatible browser. See the renderer and capture script for their environment settings. They emit render provenance alongside the output; they are not a complete recipe for fitting the historical configurations. Ordinary README builds reuse the saved saddle assets and need none of these WebGL tools.

For a fully executable current-method example with known graph/reference semantics, use the synthetic graph guide. Graph-path error, straight-line-distance error, and reference agreement answer different questions; see Choose a score.