braingrad

Essentially, braingrad is a lightweight deep learning library.

We aim to implement as many features as possible to make it capable of training basic neural network tasks.

Features:

• Tensor object
• Automatic differentiantion engine
• Forward propagation
• Some loss and activation functions
• SGD optimizer

TODO:

• Refactor autograd
• Creating sequential models
• Documentation
• nn.py:
  • use random.uniform from Tensor class
  • use activation functions defined in Tensor instead of linear lambda x:x
  • define linear activation function in engine.py (ps: x_grad = out.grad)

Contributors:

Ilyes Arfaoui

Aziz Amari