Better support for iterable datasets #3173
Merged
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
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
Seems like I've pushed some cruft in the notebooks - let me remove that and force-push before this is reviewed. |
- Fix `create_item` to be compatible with both iterable and indexed datasets. - If a sample item is required, index `do_item` with `None`, not `0`. For iterable datasets, this will get the next item, for indexible datasets it will get the first item. - Subclasses of `DataLoader` are updated to be compatible with this indexing scheme. - Native PyTorch `IterableDataset`s have a stubbed `__getitem__` method (that just raises an error), so don't just rely on the presence of that method when establishing the `indexed` property - check for `IterableDataset` classes/subclasses too.
jcaw
force-pushed
the
fix-iterable-datasets
branch
from
f170a4c
to
1c967eb
Compare
|
Looks good! - Many thanks :) |
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Outdated suggestions cannot be applied.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.
IterableDatasetobjects currently don't work when_one_passis called, because this method attempts to index item0butNonewould be required for anIterableDataset. However, passingNoneis incompatible with indexed datasets. This PR aims to fix the issues.The following changes are made:
create_itemto be compatible with both iterable and indexed datasets.do_itemwithNone, not0. For iterable datasets, this will get the next item, for indexible datasets it will get the first item.DataLoaderare updated to be compatible with this indexing scheme.IterableDatasets have a stubbed__getitem__method (that just raises an error), so don't just rely on the presence of that method when establishing theindexedproperty - check forIterableDatasetclasses/subclasses too.While I've tried to familiarise myself with how
_one_passis used, I'm a little unsure whether it could be called such that it consumed the first item of the iterator, then the iterator continued to be used by e.g. the training loop (effectively skipping the first item). Is this a problem, or is the iterator always reset by callingcreate_batches?