An Neural Network Libary, made completely from scratch using numpy and python, originally based on this article, currently supports:
- loading, and automatic saving, of trained models
- Fully Connected Layers
- Convolutional Layers
- Softmax
- Sigmoid Activation
- Tanh Activation
- ReLU Activation
- Custom layer types (inherits
Layer) - Custom Activations and Loss functions (inherits
NNFunction) - Mean Squared Error loss
- Categorical Cross Entropy loss
- update pooling layer for multiple color channels
- convert modules to C for speed?
- UNets
- Diffusion Models