Chainer v7.0.0rc1 をリリースしました

[Hiroyuki Yamazaki]

Chainer v7.0.0rc1 をリリースしました！ リリースノートは以下の通りです。

This is the release note of v7.0.0rc1. See here for the complete list of solved issues and merged PRs.

Announcements

This time, we will keep the current branches for active development (master for v7.x, v6 for v6.x) after the RC. We will maintain v6.x series until Python2 EOL, so we do not cut the new development version for now to avoid increasing the number of branches to maintain. New features will be included directly into v7 for a while, and maintenance changes will be backported to v6.

Highlights

ONNX-Chainer Integration

ONNX-Chainer which used to be a separate project has now been integrated to the Chainer repository and made more accessible to existing Chainer users (#8229). You can easily export Chainer model as ONNX format like this:

import onnx_chainer
onnx_chainer.export(chainer_model, pseudo_input, filename='model.onnx')

For a more detailed description on how to get started, please refer to the ONNX-Chainer section in the official documentation.

ChainerMN

ChainerMN now works with ChainerX. In this release, the MNIST example has also been updated to demonstrate the usage. (#7844)

New Features

• Add UpsamplingDeconvFilter and DownsamplingConvFilter initializer (#5290, thanks @knorth55!)
• Add chainerx.meshgrid (#6668, thanks @kshitij12345!)
• Add chainerx.hsplit (#7030, thanks @ishanrai05!)
• Add linalg.cholesky to ChainerX (#7329, thanks @IvanYashchuk!)
• Add linalg.eigh, linalg.eigvalsh to ChainerX (#7503, thanks @IvanYashchuk!)
• ChainerX + ChainerMN integration on MNIST (#7844)
• New configuration system of communicator inspired by links (#7885)
• More efficient multi-node snapshot (#8003)
• A new multi-node evaluator for force_equal_length=False (#8071)
• Allow weight initializer to have its own RandomState instance (#8081, thanks @mr4msm!)
• Add chainerx.hinge (#8168)
• Integrate ONNX-Chainer to Chainer repository (#8229)
• Implement chainerx::SoftmaxCrossEntropy and chainerx.softmax_cross_entropy (#8250)
• Add chainermn.testing.to_device function (#8279)
• Add chainerx.copyto (#8314, thanks @kshitij12345!)

Enhancements

• Rename TabularDataset.as_tuple/as_dict to TabularDataset.astuple/asdict (#7788)
• Deprecate DeviceResident.to_gpu/to_cpu/to_intel64 (#8058)
• Support zero-sized matrix in generate_matrix (#8167)
• Add mode argument to chainerx.take (#8197)
• Delete move and copy of virtual *GradState classes (#8224)
• Fix directional gradient stability in gradient_check (#8236)
• Fix some typo (#8243, thanks @garanews!)
• Fix CuPy installation detection error message (#8264)
• Fix intel64 support of F.batch_normalization (#8266)
• Fix dim clearing on output (#8270)
• Remove device argument from chainerx.diag and chainerx.diagflat (#8275)
• Fix algorithm to avoid small directions in gradient_check (#8290)
• Show import error with guild message on ONNX (#8293)
• Partially output_grad support on fake_as_funcnode (#8298)
• Compute F.negative_sampling in fp32 for fp16 inputs (#8300)
• Make some arguments keyword-only. Note that some of them may break code based on v7 beta versions, but none of them breaks the compatibility against v6.
  • Make mode and align_corners arguments in F.resize_image keyword-only (#8009)
  • Make weights and keepdims arguments in Variable.mean keyword-only (#8010)
  • Make arguments of WeightStandardization keyword-only (#8011)
  • Make call_before_training argument of Trainer.extend keyword-only (#8064)
    • The argument was introduced in v7.0.0b3, so it is not counted as compatibility break of v7.
  • Make arguments in ObservationAggregator and MultiNodeEarlyStoppingTrigger keyword-only (#8065)
  • Make force_equal_length argument in scatter_dataset and scatter_index keyword-only (#8066)
  • Make size argument of tabular.from_data keyword-only (#8067)

Performance Improvements

• Make contiguous case for chainerx::Take faster (#8295)

Bug Fixes

• Fix subgraph construction for ChainerX backward (#8049)
• Fix a bug in F.batch_normalization with mixed dtype (#8149)
• Fix __str__ of parameterized class (#8169)
• Fix bugs when x and gamma/beta have different dtypes in F.batch_normalization (#8175)
• Change copy to __deepcopy__ in ChainerMN batch_normalization and replace to_gpu (#8185)
• Fix possible data race in CUDA memory keeper (#8213)
• Add virtual destructor to CUDA Allocator (#8215)
• Inherit input ndarray device in chainerx.ascontiguousarray (#8262)
• Do not expose global_kernel_registry (#8265)
• Fix SCE with ChainerX and normalize (#8301)
• Unable to use gpu_id=0 in ChainerMN testing get_device (#8304)

Code Fixes

• Update variable names for consistent naming convention (#8074)
• Fix style of setup.cfg (#8180)
• Remove unused forward declaration of AveragePoolPadMode enum (#8214)
• Write Read the Docs related comments in setup.py (#8218)
• Remove unused classes {Max,Average}PoolForwardBackward (#8223)
• Conform to readability-avoid-const-params-in-decls (#8225)
• Simplify direction vector sampling in gradient_check (#8238)
• Use type hint for method declaration (#8248)
• Remove obsolete comment in F.softmax_cross_entropy (#8253)
• Fix import order and grouping (#8257)
• Simplify CreateSubgraph (#8310)

Documentation

• Change citation to new KDD paper (#7994)
• Fix a typo in the Cauchy distribution page (#8208, thanks @nzw0301!)
• Fix resize_images documentation to reflect recent code changes (#8221, thanks @zu3st!)
• Set up documentation for loss functions in ChainerX (#8231)
• Add documentation for chainerx.ravel (#8233)
• Add documentation for chainerx.sigmoid_cross_entropy (#8249)
• Put a link to CuPy installation guide in README instead of a command instruction (#8287)

Installation

• Add ability to build with ninja generator. (#8194, thanks @cloudhan!)
• Suppress warnings-as-errors from external libraries (#8227)
• Write CMake generator when building (#8239)
• Add libchainerx_base.a to link chainerx statically (#8247)

Examples

• Fix WaveNet example not working (#8157, thanks @dhgrs!)
• Fix generate.py in examples/wavenet (#8172, thanks @dhgrs!)

Tests

• Simplify F.scale test (#6969, thanks @ishanrai05!)
• Improve example tests (#7475)
• Add fp16 test to test_n_step_rnn (#7483)
• Fix protobuf dependency (#7529)
• Fix TestAccuracy: Randomly reduce testing parameters (#7820)
• Support ChainerMN testing in pfnci (#7821)
• Fix flaky tests of chx.linalg.solve (#7997)
• Fix overflow warning in div backward test (#8109)
• Fix flaky TestQR (#8114)
• Disable flaky test retry in flexCI (#8143)
• Pairwise testing (#8164)
• Allow pytest.skip() in combination with testing.repeat/retry (#8174)
• Remove DummySerializer and DummyDeserializer from iterators_tests (#8176)
• Fix comparison with casting in hdf5 serializer test (#8182)
• Relax BatchNormalization backward test tolerances (#8189)
• Fix caffe test with protobuf>=3.8 (#8190)
• Add CHAINER_TEST_PAIRWISE_PARAMETERIZATION and enable it only in Travis CI (#8211)
• Fix attrs package version (#8219)
• Fix HDF5Serializer test for h5py<2.9 (#8220)
• Fix flaky TestBatchNormalization (#8230)
• Relax tolerances in ChainerX unary math tests (#8234)
• Add "jenkins" extras (#8241)
• Use clang-format-6.0 if possible and track the version of clang-format (#8242)
• Remove legacy DeprecationWarning filter from test_multi_node_chain_list (#8246)
• Fix chainex_tests/unit_tests/routines_tests/test_linalg.py::Inverse (#8255)
• Fix flaky TestHuberLoss (#8271)
• Stop setting too small tolerances in backprop tests (#8283)
• Make ImportWarning just a warning in tests (#8291)
• Fix gtest linkage (#8292, thanks @cloudhan!)
• test_average is slow in FlexCI (#8303)
• Add ChainerX to test_mnist in chainermn_tests (#8305)
• Implement communicator_test for ChainerX+ChainerMN (#8313)

Others

• Remove ImportWarning ignore entry (#8186)
• Add WIN32_LEAN_AND_MEAN definition (#8205, thanks @cloudhan!)
• Deprecate multinode checkpointer (#8207)
• Replace Slack invitation links (#8263)