kedro-datasets-9.5.0

Major features and improvements

• Added send_individually option to APIDataset to send list items as individual requests instead of batched arrays.

Bug fixes and other changes

• Hardened experimental pytorch.PyTorchDataset: weights_only=True is now enforced by default on load to block arbitrary code execution from untrusted .pt files, user-supplied load_args and save_args are now correctly passed to torch.load and torch.save (previously silently dropped), and the misleading "pickle-safe" docstring was corrected.
• Fixed the darts-torch-model-dataset optional dependency to point at the real PyPI package u8darts[all].
• Repaired polars.PolarsDatabaseDataset end-to-end and added a full test suite for it.
• Fixed opik.TraceDataset so credentials.project_name is now passed to configure() and persisted to Opik's session configuration.
• Added os.PathLike support for Spark datasets.
• Fixed ibis.FileDataset to support remote filepaths (e.g. s3://, abfss://, hf://) and added an fs_args argument to authenticate the filesystem used for version discovery.

Community contributions

• PragnyaKhandelwal
• Anton Nikishin
• GDaamn