Repository navigation
Fix EarlyStopping using stale losses between display steps - #2109
Open
maximbeekenkamp wants to merge 1 commit into
Open
maximbeekenkamp wants to merge 1 commit into
maximbeekenkamp wants to merge 1 commit into
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #2048.
EarlyStoppingchecks the loss every iteration, but the training loop only refreshes it atdisplay_everyintervals. This can exhaust patience while the loss is still improving.With this change, attaching
EarlyStoppingevaluates losses every iteration. Printing and loss history keep their configured interval, with the stopping step also recorded. This adds evaluation cost when early stopping is enabled.Validated with analytically calculated supervised and PDE losses across all five CPU backends. All 13 PyTorch CUDA integration cases passed on an RTX 3090. Existing external-optimizer limitations remain unchanged.