A template for neural network experiments
- Define your PyTorch Dataset, your NN, and your randopt variables. Call
learn(...), and off you go. - Automatic splitting (if desired) into train/valid/test data.
- Choose to use CUDA (automatically distributed) or not.
- Use sensible - but overrideable - defaults. (for optimizer, data pre-processing, hyper-params, etc...)
- Provide Mnist example.
- train function should only do 1 epoch.
- Sensible command-line arguments support.
- Easy network definition.