Interactive WALNUTS sampling#

Runs in the browser via numbl — no install.

🎛 Just the figure →#

The figure-only view runs the sampler and drops you straight into the interactive figure (it shows the run's output first, then the figure). The full developer view — file tree, editable code, console — is below.

▶ Open walnuts_demo.m and click Run#

Samples are drawn from a 2D target by WALNUTS (the within-orbit adaptive leapfrog No-U-Turn Sampler). The figure shows the target density as a heatmap with the samples scattered on top. Switch between several targets (banana, Gaussian, correlated Gaussian, donut) to see the sampler handle different geometries.

WALNUTS is a NUTS variant that grows a Hamiltonian orbit by NUTS-style doubling until a U-turn, but within each step it refines the leapfrog step size — halving the step (and doubling the count) until the energy variation over the step is within a tolerance δ — then picks a draw from the orbit.

Controls:

• Target — switch the distribution (banana / Gaussian / correlated / donut).
• Samples / Step h / Energy tol δ — change a setting to re-run the sampler with it (a larger base step h or smaller δ triggers more within-orbit refinement).
• Resample — a fresh chain with the current settings.
• ▶ Movie — animate WALNUTS building orbits step by step: each transition traces its leapfrog path (amber), then circles the selected draw, which joins the chain and starts the next transition.

How it works#

• walnuts_demo.m — driver: addpath('helpers'), seed, call the sampler.
• walnuts_sampler.m — loops the sampler (one transition per draw, passing the target as function handles), evaluates the target on a grid, opens the figure.
• helpers/ — Nawaf Bou-Rabee's reference WALNUTS (walnuts.m + extend_orbit_*, micro, leapfrog, u_turn/sub_u_turn, p_micro/pmf_p_micro), and the targets log_density.m / grad_log_density.m.
• app/ — a single-file React app that draws the density heatmap + samples on a canvas.

The script sends the density grid + samples via uihtml(..., 'Data', ...). The controls call back: resample re-runs the chain, setTarget switches the target (and rebuilds the density), and movie records a few transitions' orbits; the script replies with sendEventToHTMLSource.

Credits#

• Algorithm & reference MATLAB implementation: N. Bou-Rabee, B. Carpenter, T. S. Kleppe, and S. Liu, The within-orbit adaptive leapfrog no-U-turn sampler, JMLR 27(113):1–64 (2026); preprint arXiv:2506.18746. The helpers/ sampler (walnuts.m + its building blocks) is Nawaf Bou-Rabee's reference MATLAB code, used here as provided. Reference C++ at flatironinstitute/walnuts.
• Target distributions (banana, donut, …) from Chi Feng's mcmc-demo.

Deploy#

Pushing to main builds the app and publishes the project to GitHub Pages via the deploy workflow.