v0.1.0

Highlights

1. Documentation is now live and a Quickstart Notebook is available
2. Lance is now integrated with pytorch and supports multiple workers.
3. Vision-specific extension types like Box2d provides vectorized iou and Image types that make it easy to perform IO and go between bytes, PIL, numpy, and tensors.

What's Changed

• Setting BatchSize via ScanBuilder by @eddyxu in #135
• Move Expression based schema project to Schema class by @eddyxu in #137
• Refactor I/O exec nodes by @eddyxu in #136
• Simplify RecordBatchReader to use Project.next() by @eddyxu in #139
• Convert bdd100k dataset in python benchmarks by @eddyxu in #131
• Fix the condition of Scan advancing batch id by @eddyxu in #143
• Initial PyTorch Dataset support by @eddyxu in #134
• Example training code over oxford pet dataset by @eddyxu in #144
• Test writing fixed size list and fixed size binary via WriteTable by @eddyxu in #151
• Fix fixed size length calculation by @eddyxu in #152
• Provide binary to profiling scans by @eddyxu in #149
• Multi-worker support in Pytorch Dataset by @eddyxu in #147
• Vision specific extension types by @changhiskhan in #146
• lance dataset that overrides Dataset.scanner and Dataset.head by @changhiskhan in #158
• Pickle Image by @changhiskhan in #160
• Only load manifest once within the dataset and share Manifest amount the readers by @eddyxu in #155
• Improve ergonomic of the Pytorch dataset and Generate embeddings for oxford pet by @eddyxu in #157
• Fix PlainEncoder to read empty page by @eddyxu in #164
• Convert coco annotations from the list of structs to struct of lists by @eddyxu in #166
• Convert coco bounding box format to [x0,y0,x1,y1] format. by @eddyxu in #169
• Image Array by @changhiskhan in #168
• Fix writing and reading extension type by @eddyxu in #172
• Coco improvements by @changhiskhan in #174
• Support partitioning and group size control in coco dataset generation. by @eddyxu in #175
• Extension type improvements to support 3d types by @changhiskhan in #173
• Support converting PIL from Image in pytorch Dataset by @eddyxu in #176
• Minor fix for 3d extension types by @changhiskhan in #177
• MS coco dataset training by @eddyxu in #163
• Change version import to relative import by @eddyxu in #181
• [python] Mix of minor improvements by @changhiskhan in #182
• Automatically build document and publish to Github Pages by @eddyxu in #180
• [benchmarks] simplify the datagen code and remove partitioning for now by @changhiskhan in #183
• Fix PlainDecoder handle empty filtered array by @eddyxu in #187
• [python] minor improvements by @changhiskhan in #190
• Fix bug that attempt to partitioned columns which does not exist in the file. by @eddyxu in #189
• Pass filter indices via Limit and Return empty array in GetListArray by @eddyxu in #191
• Exclude filter columns from projection by @eddyxu in #194
• action to bump version for new release by @changhiskhan in #199
• [C++] [BUG] Adjust offset when the batch size is set for reading by @eddyxu in #201
• GH action to upload wheels and also make reusable yml by @changhiskhan in #200
• [Python] Test projection in Python Torch Dataset by @eddyxu in #202
• Fix typo of calculating offsets for slicing index by @eddyxu in #206
• Changhiskhan/tutorial by @changhiskhan in #167

Full Changelog: v0.0.5...v0.1.0