add test if dataset samples can be collated#5233
Open
pmeier wants to merge 1 commit intopytorch:mainfrom
Open
Conversation
Contributor
💊 CI failures summary and remediationsAs of commit 18799db (more details on the Dr. CI page):
🕵️ 9 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
| Job | Step | Action |
|---|---|---|
| Run tests | 🔁 rerun |
🚧 1 ongoing upstream failure:
These were probably caused by upstream breakages that are not fixed yet.
- unittest_prototype since Jan 14 (adf8466)
This comment was automatically generated by Dr. CI (expand for details).
Please report bugs/suggestions to the (internal) Dr. CI Users group.
Member
|
Summing up what we discussed with Philip offline: The issue here is that some of our datasets (those above) can have The default collate function of DataLoader cannot handle dataset = datasets.CLEVRClassification("~/datasets", split="test", transform=transforms.PILToTensor())
data_loader = DataLoader(dataset, batch_size=3)A decent workaround for these right now is to write a custom collate function that doesn't try to wrap def collate_fn(batch):
imgs = [img for (img, _) in batch]
return torch.stack(imgs), [None] * len(batch)We'll reach out to the torchdata team to see what they think of this. Perhaps the default collate function could start accepting |
Contributor
Author
Contributor
|
QQ: If there is no target in test set, why not emit target and only keep the |
Member
|
It's because we want to keep the returned types / length consistent across splits. Otherwise users need to unpack the returned values and check whether there is a target or not, which can be a cumbersome |
Contributor
Makes sense. And, what would we expect in the batch before collate if target is
And, we need a custom map (collate) function for either case. def collate_fn(batch):
imgs, targets = tuple(zip(*batch))
return default_collate(imgs), NoneCase 2: def collate_fn(batch):
return default_collate(batch), NoneI'm using |
Contributor
Author
That is the root of the problem: we cannot say past |
Contributor
Author
|
I've summarized my proposal here. Please have a look @NicolasHug, @ejguan, @NivekT, @VitalyFedyunin |
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
4 participants
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 following datasets currently fail the test:
WIDERFaceKittiSintelKittiFlowHD1KFER2013CLEVRClassificationOxfordIIITPetThey all fail since they return
Nonein at least one configuration anddefault_collatedoes not support it. This in turn means that the datasets with the offending configurations cannot be used in atorch.utils.data.DataLoader(dataset)without passing a customcollate_fn.cc @pmeier