A quick implementation that I think is more or less working. The idea is to partition off 5% of the data for validation to check for overfitting during training, and also perhaps as a more deterministic metric across runs when testing hyperparameters.
Also snuck in a separate commit that removes .pyc files from the repo.
A quick implementation that I think is more or less working. The idea is to partition off 5% of the data for validation to check for overfitting during training, and also perhaps as a more deterministic metric across runs when testing hyperparameters.
Also snuck in a separate commit that removes .pyc files from the repo.