autograd

Why we included this project

Autograd brings automatic differentiation to plain NumPy. You write the forward pass as ordinary Python and NumPy with the usual control flow, and the library returns a function that computes gradients. It supports both reverse-mode (backpropagation) and forward-mode, and the two compose, which covers most gradient-based optimization work in research and prototyping without dragging in a heavyweight framework. Since it can keep differentiating, higher-order derivatives come free, which matters for some optimization and sensitivity analyses. It also works well as a teaching tool: the README's tanh example makes backpropagation concrete in a few lines. For anyone already comfortable in the NumPy ecosystem, it is a pragmatic way to add derivatives without changing how you write math.