Phase 3 - CR vs CR/PR | Single Domain
Created on October 3|Last edited on October 11
Comment
Main Experiment
Run set
5
Fixing NaNs
This happens down the road. Still uncertain as to why since i couldn't see a snapshot of the model before the event but I've started another run with same hyperparameters except I'm using the old BCE loss. If this converges without a loss in performance, we're golden and should carry on with the BCE loss in trying to get pruner to work well.
Run set
2
Add a comment