Sampling Safari MCMC Trails

Follow chance. Find every landmark.

Each blue landscape is a probability distribution. Your sampler wanders through it, leaving a trail. Collect every numbered gold landmark by landing close to it.

1. Choose a field and a travel method, then Start exploring.
2. Tune the sampler while it runs. Tiny jumps can get stuck; huge jumps may be rejected. Momentum can travel farther, but needs a sensible integration step.
3. Green arrows mean accepted moves. Red arrows mean rejected proposals: the sampler stays where it was. The blue edge histograms show where it has spent time.
4. Pause whenever you like. One step advances one animation event; a sample proposal can take several events. Fast trail skips the proposal animation.
5. Collect all the landmarks to earn a stamp. Fewer samples earn a brighter medal. New fields, methods or Fresh expedition reset the current trail; parameter tuning keeps it.

Randomness means every expedition differs. Landmark collection is a playful exploration goal, not a convergence test. Your passport stays on this device.

Open trails

Source: github.com/chi-feng/mcmc-demo · 948 stars observed November 7, 2026 · MIT · maintained August 28, 2026.

Original sampling algorithms, target mathematics, linear algebra and contour/histogram renderer by Chi Feng, © 2016. Contour library by Jason Davies and credited predecessors; original permissive notices are bundled. Safari interface, landmarks and passport added for this static edition. No external assets or accounts.