Modify Workflow to Allow IterableDataset Inputs #8263
+23
−10
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
Signed-off-by: Eric Kerfoot <eric.kerfoot@kcl.ac.uk>
for more information, see https://pre-commit.ci
Signed-off-by: Eric Kerfoot <eric.kerfoot@kcl.ac.uk>
…gine_stream_fix Signed-off-by: Eric Kerfoot <eric.kerfoot@kcl.ac.uk>
Signed-off-by: Eric Kerfoot <eric.kerfoot@kcl.ac.uk>
|
/build |
KumoLiu
approved these changes
Dec 16, 2024
| raise ValueError("If data_loader is not PyTorch DataLoader, must specify the epoch_length.") | ||
| try: | ||
| epoch_length = len(data_loader) | ||
| except TypeError: # raised when data_loader is given an iterable dataset which has no length |
There was a problem hiding this comment.
Do you think we should add a warning here or validate the dataset instance in the data_loader?
isinstance(data_loader.dataset, Iterable)
There was a problem hiding this comment.
The data_loader argument itself isn't always a DataLoader instance so may not have a dataset member, it can be a Iterable itself which might have a length. It think there's a use case where you'd pass in just a list of items you already have a so not use a DataLoader at all and so this is the more robust way of attempting to get a length.
There was a problem hiding this comment.
So it looks like if epoch_length is None: this if should be moved to the outside level, right?
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Outdated suggestions cannot be applied.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.
Description
This modifies the behaviour of
Workflowto permitIterableDatasetto be used correctly. A check against theepoch_lengthvalue is removed, to allow that value to beNone, and a test is added to verify this. The length of a data loader is not defined when using iterable datasets, so try/raise is added to allow that to be queried safely. This is related to my work on the streaming support, in my prototype gist I had to provide a bogus epoch length value in the then change it toNonelater once the evaluator object was created. This PR will remove the need for this hack.Types of changes
./runtests.sh -f -u --net --coverage../runtests.sh --quick --unittests --disttests.make htmlcommand in thedocs/folder.