Nerve Discover

symbolic regression · webassembly

Give it numbers. Get back an equation.

Not a prediction, not a set of weights — a formula you can read, check against theory and evaluate forever after in nanoseconds. There is no model to download here: equation discovery is pure algorithm, so the whole engine is a WebAssembly module of a few tens of kilobytes. Your data never leaves this tab.

1 · the data

Inputs in the first columns, the value to explain in the last. A header row names the variables.

2 · what it may use

Fewer operators, smaller haystack. Keep it algebraic unless the thing you are modelling really oscillates or decays.

3 · the search, live

generation 0 / 0 normalised error nodes