Chainer v7.0.0b4 をリリースしました。

[Emilio Castillo]

Chainer v7.0.0b4 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 v7.0.0b4. See here for the complete list of solved issues and merged PRs.

Highlights

Many updates to ChainerX including new routines and support for loss scaling.

New Features

• Support all float dtypes in F.n_step_rnn and F.n_step_birnn (#5808)
• Add chainerx.vsplit to ChainerX (#7032, thanks @ishanrai05!)
• Add chainerx.linalg.qr to ChainerX (#7379, thanks @IvanYashchuk!)
• Add chainerx.accuracy (#7526, thanks @aksub99!)
• Add chainerx.{remainder/mod} (#7675, thanks @sky58!)
• Add Tree-LSTM to ChainerX (#7720, thanks @dido1998!)
• Add S-LSTM to ChainerX (#7783, thanks @dido1998!)
• Loss scale support for chainerx (#7979)
• Add F.zeta (#8059, thanks @UmashankarTriforce!)
• Add testing.generate_matrix to get matrices of given singular values (#8077)
• Add chainerx.fmod (#8110)
• Add chainerx.nonzero (#8124)

Enhancements

• Abbreviate output of chainerx::ArrayRepr for large inputs (#7708)
• Make parameterized test names deterministic (#7945)
• Raise FutureWarning on GPU-to-GPU transfer in StandardUpdater (#7952)
• Always get typeid of kernels in libchainerx (#7970)
• Fixed support of 0-sized arrays for linalg routines in ChainerX (#7980, thanks @IvanYashchuk!)
• Support CuPy/ChainerX arrays when initializing variable.Parameter objects (#8022)
• Add cuda ScanKernel (#8103)

Performance Improvements

• Add chainerx::Absolute device implementation (#7319)
• Make weight standardization faster (#7963)

Bug Fixes

• Fix deadlock on MultiprocessIterator and MultiprocessParallelUpdater (#7511)
• Support mixed16/float16 GroupNormalization (#7965)
• Change return policy for chx::Device object on ndarray pickling (#7988)
• Fix deepcopy for chain parameters (#7996)
• Fix floating point exception in ChainerX inferred reshape (#8018)
• Fix chainerx::Dot edge cases with empty arrays (#8020)
• Fix LSTM for omitted upstream gradients (#8037)
• Fix native AddAt implementation for float16 arrays (#8055)
• Correctly cast fill_value in constant initializer (#8089)

Code Fixes

• Simplify ArrayReprImpl (#7699)
• Remove unnecessary file (#8000)
• Refactor F.batch_normalization and ChainerMN backend implementations (#8039)
• Fix -Wabsolute-value for clang (#8045)
• Generalize and simplify NativeCumsumKernel (#8053)
• Fix coding style of some imports in ChainerMN (#8060)
• Fix -Wbraced-scalar-init for clang (#8076)
• Use standard constructor (#8088, thanks @cloudhan!)
• Remove unused headers in arithmetic.{h,cc} (#8128)

Documentation

• Fix doc of backend.copyto (#7832)
• Document chainerx.to_numpy (#7984)
• Fix RNN docs for ChainerX (#7985, thanks @dido1998!)
• Remove obsolete note about chainerx.take indices dtype (#7998)
• Add undocumented arguments to snapshot extension signature (#8004)
• Fix grammatical errors in documentation (#8029)
• Fix heading anchor in ChainerX docs (#8091)
• Documentation improvement CHAINERX_ENABLE_{BLAS,LAPACK} (#8099)
• Add document print runtime info (#8125)
• Fix RNN documentation (#8144)
• Add documentation for chainerx.minimum (#8146)
• Remove obsolete note in chainerx.maximum doc (#8147)
• Fix typo (#8160)

Installation

• Fix NumPy version in Dockerfile (#8027)
• Add cblas.h and modified CMakeLists.txt (#8052, thanks @okdshin!)
• Fix environment variable CHAINERX_ENABLE_LAPACK=0 causes error (#8086, thanks @cloudhan!)
• Update abseil to new release (#8120)

Examples

• Use some latest features for the WaveNet example (#6285)
• Separate training script into main part and data submodule to avoid an error related to NVIDIA DALI. (#8127, thanks @lazykyama!)

Tests

• Treat warnings as errors in tests (#6653)
• Filter DeprecationWarning in test_maniplation.py (#7824)
• Avoid unnecessary test condition in F.max_pooling_2d test (#7924)
• Add test for optimizers test coverage (#7927)
• Fix flaky negative_sampling (#7975)
• Avoid testing full combinations in F.lstm test parameterization (#7987)
• Relax tolerances in gradient_check test (#7989)
• Drop Python 2 Travis CI configuration (#8013)
• Drop Python 2 AppVeyor configuration (#8014)
• Drop Python 2 PFN CI configuration (#8017)
• Suppress number of combinations of in_out_dtype (#8023)
• Avoid non-differentiable point in min/max tests (#8044)
• Adjust TrueDiv tolerances (#8047)
• Add scripts for Docker base images for Chainer CI (#8075)
• Fix tests of L.BatchRenormalization and adjust tolerances (#8080)
• Add timestamp to Travis CI log (#8085)
• Explicit h5py.File mode (#8090)
• Fix flaky tests with np.empty (#8096)
• Revive clang-tidy test in Travis CI (#8098)
• Fix matrix generation in linear algebra PseudoInverse test (#8102)
• Remove duplicated parameter in test_normal.py (#8111)
• Register pytest markers (#8112, #8132)
• Fix macOS Travis error caused by Homebrew (#8115)
• Add ignore::ImportWarning to setup.cfg (#8131)
• Relax tolerance of im2col test (#8133)
• Allow fix_random decorator to be used with OpTest (#8136)
• Fix missing dtype checks in ChainerX loss test (#8141)
• Fix flaky NStepRNN and NStepBiRNN (#8142)
• Avoid empty in F.cast test that can cause overflow warning (#8152)
• Make xdist usable in ChainerX tests (#8155)
• Adjust TestConvolution2DFunction::test_double_backward fp16 tolerance (#8163)

Others

• Convert tabs to spaces in setup.cfg (#8154)