Model checkpointing is done by validation loss, not validation accuracy

allenai / deep_qa

A deep NLP library, based on Keras / tf, focused on question answering (but useful for other NLP too)

Apache License 2.0

404 stars 133 forks source link

Model checkpointing is done by validation loss, not validation accuracy #199

Closed matt-gardner closed 7 years ago

matt-gardner commented 7 years ago

This means that the "best epoch" as defined by our code may not have actually been saved by Keras. We need to be sure that the model checkpointing uses the same metric as early stopping.