Tiny-tensorgrad

A tiny deep learning framework using numpy as backend. It has a tensor autograd engine inspired by Andrej Karpathy's micrograd and a pytorch like API

Why another deep learning framework?

There's no need to for a new deep learning framework but it is useful as a learning tool.

Demo

The file demo.ipynb has a demo classification problem. And in demo-mnist.ipynb a demo of classifying MNIST digits.

Test

Run test against pytorch

python test.py

TODO

• Add sigmoid, softmax and train with cross entropy
• Add convolutional layers
• Test on MNIST
• Add optimizers