autodiff

This is a simple eager mode automatic differentiation engine written in Python working on scalar type (Python's Real Number).

The motivation to write this project is to give myself a solid understand of how automatic differentiation works in training a neural network. This project also is also good for education to better understand the rationale behind the autodiff.

Usage

import autodiff as ad

x1, x2 = ad.Var(2), ad.Var(3)
y = x1 * x2

y.backward()

print(x1.grad, x2.grad)

>>> 3, 2

Requirements

Python 3.6+