Upgrade torchmetrics #2017
Merged
Upgrade torchmetrics #2017
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
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
… into upgrade-torchmetrics
dakinggg
reviewed
Mar 1, 2023
dakinggg
approved these changes
Mar 1, 2023
|
V11 Pass V10. Added |
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Outdated suggestions cannot be applied.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.
What does this PR do?
Upgrades torchmetrics to support 0.11.x. This is not straightforward because torchmetrics.Accuracy() now requires a task argument (either 'binary', 'multiclass', or 'multilabel'). This PR replaces all instances of torchmetrics.Accuracy() in Composer with multiclass Accuracy, with the correct number of classes provided. torchmetrics.MatthewsCorrCoef() is also updated similarly.
Because torchmetrics.Accuracy(task='multiclass') requires a num_classes argument, we add a new, optional 'num_classes' argument to ComposerClassifier. Users must either
(1) Pass in num_classes to the ComposerClassifier, or
(2) Specify a num_classes parameter in the PyTorch network submodule.
This is because if we are to use Accuracy as our default training and validation metric, we have to know how many classes the user wants. Alternatively, if users do not want to use our default training/validation metric, they can
(3) Pass in both train_metrics and val_metrics to Composer Classifier
and we will not require
num_classessince we no longer need to create a default metric.If none of these three options are satisfied, ComposerClassifier will now raise an error.
The rest of the changes in this PR are fixing the various models, tests, notebooks, and documentation snippets that depend on ComposerClassifier and properties of its Accuracy metric, such as the class name or shape.
What issue(s) does this change relate to?
CO-1836