Highlight
- Provide a new set of API for Tensorflow and PyTorch. It is more intuitive and user-friendly.
- Support FlinkML estimator style API in Python, you can check out the quick start examples for Tensorflow 1.15 and PyTorch.
- Support iterative training with data from Flink table.
- Dependencies management of Python API is now seamlessly integrated with the dependencies management of PyFlink, detail can be found here.
What's Changed
- Bump dl-on-flink version by @Sxnan in #700
- Update quick start doc by @Sxnan in #695
- Bump hadoop-common from 2.8.0 to 2.10.1 by @dependabot in #697
- Bump junit from 4.12 to 4.13.1 by @dependabot in #688
- Bump httpclient from 4.5.6 to 4.5.13 by @dependabot in #684
- Recommend using virtualenv for quick start by @Sxnan in #703
- Rename modules from flink-ml-* to dl-on-flink-* by @Sxnan in #702
- Add inference example to quick start by @Sxnan in #705
- Run example in github action nighlty by @Sxnan in #707
- Include pybind as submodule by @Sxnan in #708
- Add github workflow to prepare release by @Sxnan in #709
- Remove extra space of commands in quick start by @Sxnan in #710
- Increase timeout waiting for AM in testing by @Sxnan in #712
- Fix unstable unit test AppMasterServerTest#multiRegisterNode by @Sxnan in #714
- Use the archive url to download flink binary by @Sxnan in #715
- Add checkstyle and spotless by @Sxnan in #718
- Refactor RoleUtils and TFUtils API by @Sxnan in #716
- NodeUtils set python path properties with entry file path at first by @Sxnan in #721
- Install dev wheels for nightly build by @Sxnan in #722
- AMInputFormat wait for am node finish before return from reachedEnd by @Sxnan in #724
- Support Tensorflow on Flink iterative batch training by @Sxnan in #723
- Add Python TFUtils API by @Sxnan in #726
- Update TensorBoardPythonRunner by @Sxnan in #730
- Update DL on Flink Tensorflow Python API by @Sxnan in #731
- Use RowCSVCoding as TF default encoding/decoding class by @Sxnan in #732
- Run python unit test in GitHub workflow by @Sxnan in #734
- Support pyFlink env configuration by @Sxnan in #733
- Bump Supported Tensorflow version to 2.4.x by @Sxnan in #736
- Update Project Copyright Header by @Sxnan in #737
- Introduce the New PyTorch API by @Sxnan in #735
- Rename PytorchContext to PyTorchContext by @Sxnan in #738
- Update quick start examples by @Sxnan in #739
- Fix prepare release workflow by @Sxnan in #741
- Specify protobuf version for dl-on-flink-tensorflow wheel build by @Sxnan in #742
- DL on Flink support FlinkML API by @Sxnan in #743
- Fix examples by @Sxnan in #745
- Avoid tight loop in NodeOperator#onEpochWatermarkIncremented by @Sxnan in #747
- Support adding extra cluster config properties to estimator by @Sxnan in #746
- Bump fastjson from 1.2.49 to 1.2.83 by @dependabot in #748
- Add flink ml estimator examples by @Sxnan in #750
- Update NodeOperator and NodeInputFormat close timeout by @Sxnan in #749
- Fix unstable failover test by @Sxnan in #751
- NodeServer#stopMLRunner should not swallow the InterruptedException by @Sxnan in #752
- Add Java examples for Tensorflow and PyTorch by @Sxnan in #753
- Add DL on Flink API docs by @Sxnan in #754
- NodeServer is interrupted periodically if it is not terminated on time by @Sxnan in #755
- Specify grpcio version upper bound by @Sxnan in #756
Full Changelog: 0.4.0...0.5.0-rc0