I will organize the messy project in next 2 months.

environment

structure

1. config.py: parameters of our models
2. predict.py: load a trained model, predict for a sample
3. state: use pickle to save something generated by our model, like the model states, datasets

run

1. python train.py to train the model