chainrule

Install

pip install git+https://github.com/jmweb-org/chainrule

If you use uv, also as an isolated CLI:

uv tool install git+https://github.com/jmweb-org/chainrule

What it does

Backpropagation is one of those things that looks obvious once you have implemented it and completely opaque before. Reading someone else's _backward closure is not the same as writing your own and finding out, from a wrong gradient, exactly which line had the sign flipped. chainrule is the small version of that exercise: a Value type whose forward and backward pass fit on one screen, plus the two tools you actually want while debugging one — a picture of the graph, and a numerical check on the gradient it produced.

In action

$ chainrule gradcheck
check        max rel. error  status
polynomial         2.43e-10  pass
activations        2.27e-11  pass
division           1.44e-10  pass
tiny_mlp           6.17e-11  pass

Features

• —A Value type whose forward and backward pass fit on one screen.
• —graph() draws the graph after backward(), collapsing nodes reused more than once.
• —gradcheck compares the analytic gradient against a central finite difference.
• —A pure-Python MLP, no tensors or batching, so what you read is exactly what trains.

Other tools