Retrained time distributed model with lower learning rate starting at epoch 3 #232
Conversation
This file contains 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
✅ Deploy Preview for silly-keller-664934 ready!
To edit notification comments on pull requests, go to your Netlify site settings. |
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## master #232 +/- ##
======================================
Coverage 87.2% 87.2%
======================================
Files 28 28
Lines 1961 1961
======================================
Hits 1710 1710
Misses 251 251
|
pjbull
approved these changes
Sep 26, 2022
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Outdated suggestions cannot be applied.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.
The time distributed model was retrained since the StepLR did not run as intended, meaning the learning rate was not cut in half at epoch 3 in the prior run. Using a lower learning rate at this point is also useful since this is the epoch at which the backbone is unfrozen with a learning rate of 1/100 of the head.
This yields to a 4 point gain in macro f1 and a slight bump in top 1 accuracy as well (which is smaller given that the improvement comes for less common species, not blanks).
This model will replace the previous time distributed model.
Additional notes:
auto_lr_find(with a workaround, not onmaster). This yielded 0.001096, essentially the same learning rate as the default (0.001). This is reassuring in that the default seems to generalize well for various batch sizes at least when training from scratch.