Toy neural network library implementation (nnlib) I used to explain automatic-differentiation implementation in a machine learning lecture at IIT Gandhinagar.
All the source of toy neural-network library is in nnlib directory. Within nnlib directory, the layers directory has files defining classes for ReLU layer and Fully-Connected layer. The losses folder has a file defining MSE loss. The neural_net.py file has definition of NeuralNet class. This class calls upon the methods of other classes to perform forward and backward propagation.
The test.py file defines and trains a toy fully connected neural network on sample dataset for a regression task.