🧠What's Changed - Support [biomedical data types - dicom and nifti](https://docs.deeplake.ai/latest/api/types/#deeplake.types.Medical) - Support [point data type](https://docs.deeplake.ai/latest/api/types/#deeplake.types.Point) - Ability to run index search on virtual columns - 2x speedup in inverted index generation
4.1.4
Fixed Bugs
- Precondition query downloads additional columns data. - Fixed credentials update after expiration - Fix the speed and memory usage of clustered index for large result sets \(10k\). - Fashionpedia convert to v4 errors out because of list htype - Fixed MAXSIM in ascending order query
Added features
- Support [Point](https://docs.deeplake.ai/latest/api/types/#deeplake.types.Point) type - Added [from_coco](https://docs.deeplake.ai/latest/api/schemas/#from_coco) to automate coco-like datasets ingestion - Partially indexed columns support in query - Better error handling in embedding ingestion - Add python api for disabling/enabling index creation
4.1.3
- Fixed MAXSIM search for different shape embeddings
4.1.2
- 100x search time improvement with MAXSIM for ColPali using pooling - Fixed deadlock issues with multiprocessing - Improved core engine stability and performance
4.1.1
- Bug fixes in `deeplake.convert` and performance improvements in data ingestion. - Fixed TQL UNION expression for the same source. - Better support for multiprocessing.
4.1
- Ability to add linked rows for images, segment masks and binary masks - Ability to save query views through tagging - Integration with MMDetection and MMSegmentation - Autocommit the data at the session exit - Ability to create dynamic arrays - Ability to create structures with known schema