search

MadryLab / robustness

A library for experimenting with, training and evaluating neural networks, with a focus on adversarial robustness.
MIT License
903 stars 181 forks source link

Include `is_best` in `LOGS_SCHEMA` #94

Open ymerkli opened 3 years ago

ymerkli commented 3 years ago

I think it would be useful to include the is_best variable from train.train_model in the logs. Since log_info is passed to the epoch hook, the epoch hook function could then perform custom model checkpointing based on whether the current model is the best model so far. For instance, I'm trying to checkpoint the classifier network itself (i.e. only AttackerModel.model, not the entire AttackerModel). This would be easier if an epoch hook had the information on whether the current model is best.

andrewilyas commented 3 years ago

This is a great idea, will make sure it's there in the next release, assuming it doesn't break anything :)