During training
If your computer breaks, you can pass it a --resume argument next time:
python run-gqn.py --data_parallel=True --batch_size=144 --workers=2 --resume=343000 (or whatever the last checkpoint number was)
You might be asking: What do the graphs mean? We only log ELBO and KL at the moment, but we should probably log more metrics! to get a better picture of how our neural net is doing