Fix chunk casting and schema unification in dataset #7873
+53
−3
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Outdated suggestions cannot be applied.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.
Updated chunk handling to cast to expected schema when features are provided or to unify schemas when not. This ensures proper schema alignment for the yielded batches.
fixes #7872
This PR fixes a bug where
IterableDatasetcreated from a generator with explicitfeaturesparameter would fail during arrow operations (like.to_pandas()) when the data contains missing or null values.Problem
When an
IterableDatasetis created with explicit features but the generator yields data with missing values (e.g., empty lists), PyArrow would infer different schemas for different batches based on the actual data rather than using the provided schema. This causedArrowInvaliderrors when trying to concatenate batches with mismatched schemas.Example error:
Solution
Modified
RebatchedArrowExamplesIterable._iter_arrow()to:This ensures that the user-provided schema is respected throughout the iteration process.
Testing
Verified the fix with the following test case:
Before Patch -
After Patch -