Explore a precomputed state space in the browser

The notebook precomputes a discrete state space of 88 combinations of function, layer scheme, and noise level. Caching those results lets you explore the whole space in a live embed — every pick served from cache, no torch required — and share it as a single static page.

Pick a function, a layer scheme, and a noise level; the fit scatter (raw data, classic NN, and KAN) updates from a cache hit. A product() sweep warms every entry of the state space at export time, so any selection is an instant restore and only the plot re-runs.

The frame embeds the static marimo export html-wasm --execute build, deployed alongside the book. The notebook is KANNS_precompute.py; every dropdown/slider pick is served from the bundled cache.