CnnLearn

A repository for implementing neural networks from scratch or notes about NN knowledge.

How to train

See example/ for example training scripts

• Syntax:
  • python -m example.train_mlp

Future features:

• Activation
• Optimizer
• Fully connected
• Convolution
• Pooling
• dataset(iter)
• Unit test
• Random generator
• Transpose
• Alex net, VGG...(basic cnn)
• Design model other than sequential
  • Probably concat
• Model zoo(Resnet, Inception, EfficientNet)
• Checkpoints
• Model training benchmarks(Cost time, accuracy...)
• Optimization
  • Vectorized derivative of softmax
• Netron visualize