v0.1.15 Support Apache Spark 3.3.1

What's Changed

• Use the same Python interpreter for drive and executor by @da-tubi in #708
• Allow 3rd party flavor to provide generate_udf by @da-tubi in #709
• Support for Spark 3.3.1 by @da-tubi in #710
• Fix unit test for Spark 3.3.1 by @da-tubi in #713

Full Changelog: v0.1.14...v0.1.15

v0.1.14 Update Spark version and relax antlr python dependency

special thanks to @dnnspark

What's Changed

• [no ci] Notebook: update Mojito for rikai 0.1.13 by @da-tubi in #700
• [no ci] Update MojitoVideo for Rikai 0.1.13 by @da-tubi in #701
• Update default Spark version to 3.2 by @changhiskhan in #705
• Relax antlr python dependency by @eddyxu in #707

Full Changelog: v0.1.13...v0.1.14

v0.1.13 scaling images

scale an image to 3 times the width and height with just img * 3

What's Changed

• torchvision >= 0.10.0 by @da-tubi in #691
• Lazy torchvision model types by @da-tubi in #693
• DSL to scale the Image by @da-tubi in #694
• Revert "Remove support in python3.7 (#670)" by @da-tubi in #695
• init rikai session with num_cores by @darcy-shen in #698

Full Changelog: v0.1.12...v0.1.13

v0.1.12 serde bug fixes

What's Changed

• Fix pandas writer bug by @changhiskhan in #686
• fix issue 640 by @Renkai in #662
• Fix pandas writer issue by @changhiskhan in #687

Full Changelog: v0.1.11...v0.1.12

v0.1.11 Extracting embeddings

Add a pytorch ModelType to extract embeddings

What's Changed

• Remove support in python3.7 by @eddyxu in #670
• A Pytorch ModelType for feature extraction and embeddings by @eddyxu in #671
• Make FeatureExtractor compatible to TorchScript by @eddyxu in #675
• Fix bug converting array type by @changhiskhan in #674
• Revert ""Bump version for python package: 0.1.11.dev0 -> 0.1.11"" by @eddyxu in #678
• bump github action to use python 3.8 by @eddyxu in #677

Full Changelog: v0.1.10...v0.1.11

v0.1.10 Write Rikai without live Spark session

Plus:

• Scala UDF to contruct Image
• Better ergonomics drawing labels

What's Changed

• Add the ability to write a pandas dataframe to disk in Rikai format without needing a live spark session by @changhiskhan in #653
• [no ci] Disable tests tagged with Network by @da-tubi in #656
• Release assembly jar by @da-tubi in #625
• [no ci] Separate RegistryTest out by @da-tubi in #658
• scala UDF image to construct a Image UDF from a URI by @da-tubi in #648
• Box2d label with Draw by @Renkai in #650
• Change the return value of mask type to string by @eddyxu in #661
• Support loading sklearn models using fs registry by @eddyxu in #665
• Serialize width/height with polygon by @eddyxu in #667
• Fix FileRegistry missing base_dir, if the model URI is local path by @eddyxu in #668
• bump python version for release job by @changhiskhan in #669

Full Changelog: v0.1.9...v0.1.10

v0.1.9 easier model inference and dataset improvements

Model registry can load models directly using a weights file now.
Fixed bug in handling Parquet conversions in nested array of structs.
Add method to retrieve number of rows in a rikai dataset

What's Changed

• withOrigin work around for DBR 10.4 LTS by @da-tubi in #637
• Use user-space pytest to avoid installing pytest again by @da-tubi in #641
• Filesystem registry directly loads model weights by @eddyxu in #642
• Update docs for SQL ML and Model Type by @eddyxu in #644
• Pin antlr version by @eddyxu in #646
• Allow to customize collate_fn per model type by @eddyxu in #643
• Fix: set options for DummyModelSpec by @da-tubi in #647
• fix dataset conversion for nested structs by @changhiskhan in #651
• Get dataset size by @eddyxu in #652

Full Changelog: v0.1.8...v0.1.9

v0.1.8

What's Changed

• Implement model types for official pytorch models by @eddyxu in #606
• Handle empty inputs in Box2d.ious by @eddyxu in #614
• Assembly dependencies cleaning by @da-tubi in #594
• [no ci] Minor fix for error message by @da-tubi in #618
• Model types for Sklearn by @eddyxu in #619
• Delete deprecated model attributes: preprocessor, postprocessor by @eddyxu in #609
• CI: cache official pretrained Torch models by @da-tubi in #623
• Make label / candidates list to be one metadata of model type by @changhiskhan in #617
• Change height and width order in parameters by @eddyxu in #628
• Do not throw ValueError when drawing image without annotations by @eddyxu in #631
• Support partitionBy with Rikai dataframe writer in Spark by @eddyxu in #635

Full Changelog: v0.1.7...v0.1.8

v0.1.7 Minor improvements and Spark version update

What's Changed

• bump to Apache Spark 3.2.1 by @da-tubi in #586
• Minor: set proper assembly name by @da-tubi in #587
• Clean up ModelType interface and small fixes by @eddyxu in #593
• Upgrade to enableIf 1.1.8 by @da-tubi in #595
• Allow set parquet block size via option by @eddyxu in #596
• Raise exceptions if files or columns not found by @eddyxu in #601
• Persist rikai specific metadata in datasets by @eddyxu in #600

Full Changelog: v0.1.6...v0.1.7

v0.1.6 user experience improvements

Minor improvements across the stack to make the user experience better with Rikai

What's Changed

• Use model name to be the default column name for ML_PREDICT by @eddyxu in #576
• Tensor dataset support2 by @Renkai in #574
• Support offset in Rikai parquet dataset by @eddyxu in #581
• Model Type with Pretrained Mixin by @da-tubi in #579

Full Changelog: v0.1.5...v0.1.6