Chainer v7.0.0a1 をリリースしました

[Akifumi Imanishi]

Chainer v7.0.0a1 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 v7.0.0a1. See here for the complete list of solved issues and merged PRs.

Highlights

• Many examples including ImageNet, DCGAN and VAE supportstart supporting ChainerX arrays

New Features

• Support orthogonal embedding initialization (#6031)
• Add an option in links.loss.CRF1d to automatically sort the input sequence (#6351)
• Add AdaBound (and AMSBound) (#6388, thanks @hitsgub!)
• Add squared_difference to chainerx (#6501, thanks @aksub99!)
• Implement array vs array functionality to chainerx.minimum (#6541, thanks @aksub99!)
• Add FP16 support to send/recv (#6552)
• Implement array to array functionality to chainerx.maximum (#6570, thanks @aksub99!)
• Add Mean Var Python Bindings to ChainerX (#6640, thanks @kshitij12345!)
• Add chainerx.ceil (#6705, thanks @kshitij12345!)
• Add chainerx.floor (#6707, thanks @kshitij12345!)
• Add chainerx.absolute (#6715, thanks @dido1998!)
• Add chainerx.argmin and chainerx.ndarray.argmin (#6740, thanks @Harshan01!)
• Add chainerx.amin and chainerx.min (#6752, thanks @Harshan01!)
• Add chainerx.a/sinh,chainerx.a/cosh (#6776, thanks @kshitij12345!)
• Add chainerx.fabs and chainerx.sign (#6777, thanks @kshitij12345!)
• Add chainerx.logical_and chainerx.logical_or (#6779, thanks @kshitij12345!)
• Add chainerx.all and chainerx.any (#6781, thanks @kshitij12345!)
• Add chainerx::Softmax and chainerx.softmax (#6814, thanks @tohmae!)
• Add zero fill mode in allreduce of chainermn (#6817)
• Make BatchNorm states public (#6847)
• Introduce Native/CUDA macros for registering standard elementwise ops (#6870, thanks @kshitij12345!)
• Make adam variants more accessible (#6874, thanks @crcrpar!)
• Add chainerx::Swapaxes and chainerx.swapaxes (#6897, thanks @kshitij12345!)
• Add chainerx.logical_xor (#7014, thanks @ishanrai05!)
• Add chainerx.log10 (#7015, thanks @ishanrai05!)
• Add chainerx.isfinite (#7016, thanks @kshitij12345!)
• Add bitwise ops to ChainerX (#7017, thanks @kshitij12345!)
• Add chainerx.arctan2 (#7028, thanks @kshitij12345!)
• Add chainerx.expand_dims (#7029, thanks @kshitij12345!)
• Add chainerx.flip, chainerx.fliplr and chainerx.flipud (#7065, thanks @kshitij12345!)
• Add chainerx.where (#7067, thanks @kshitij12345!)
• Add F.arctanh (#7095)

Enhancements

• Improve error message of gradient_check.check_double_backward (#6427)
• Improve link_hooks.SpectralNormalization (#6655, thanks @crcrpar!)
• ChainerX Op registration: normalization (#6719)
• ChainerX Op registration: arithmetic (#6723)
• Implement Relu in ChainerX (#6731, thanks @dido1998!)
• Make device functions public (#6744)
• ChainerX Op registration: creation (#6745)
• ChainerX Op registration: linalg (#6746)
• Allow snapshot_object have condition and writer option (#6762)
• Support fallbacks of ChainerX on GetItem fail when indices contains chainerx.ndarray (#6769)
• Fix Evaluator for chainer.dataset.converter (#6768)
• Rename patients argument to patience in EarlyStoppingTrigger (#6784)
• Remove Backend ctor and use CreateBackend (#6785)
• ChainerX Op registration: pooling (#6800)
• Define __str__ for Device classes (#6816, thanks @nishnik!)
• Simplify numeric.h (#6832)
• ChainerX Op registration: connection (#6833)
• ChainerX Op registration: array members (#6834)
• ChainerX Op registration: math (#6842)
• Mixed dtypes: chainerx::Minimum (#6858)
• Update distributions.independent (#6860, thanks @ganow!)
• Add chainerx.ndarray.all and chainerx.ndarray.any (#6926)
• Fix HuberLoss.forward avoid loss of significance (#6940)
• Support Tensor Core in chainerx::Dot (#6960)
• Fix F.get_item backward for ChainerX (#6991)
• Support NumPy scalars in ChainerX arithmetics (#7004)
• Implement NumPy-like pairwise reduction for stability (#7043, thanks @grafi-tt!)
• Support mixed dtypes in Stack (#7058)
• ChainerX Scalar / Array divisions (#7075)
• Fix Reshape copy condition (#7080)
• Fix trigger constructors to raise errors instead of assertion failures (#7101)
• Support Tensor Core in chainerx::Conv (#7112)

Performance Improvements

• Optimized ChainerX-to-CuPy ndarray conversion (#6204)
• Use cuDNN in ReLU (#6993)
• Fast integer scale unpooling (#7114, thanks @tkerola!)

Bug Fixes

• Avoid throwing in destructors (#6725)
• Fix TypeError during BN deserialization on Win64 (#6765, thanks @hyabe!)
• Fix chainerx.astype casting from float16 to bool in CUDA (#6780, thanks @kshitij12345!)
• Fix ArgMax of CUDA when all values are negative (#6783)
• Fix unchain gradient pull (#6804, thanks @Rishav1!)
• Remove chainerx.square fallback since it is implemented in C++ (#6823)
• Fix stack overflow caused when to_gpu/to_cpu/to_intel64 were overridden (#6824)
• Fix filename arg of PlotReport (#6866)
• Make InvalidType picklable (#6884, thanks @zaltoprofen!)
• Rename the macro name for AMinOp (#6922)
• Fix terminal column width retrieval in backprop traceback in Python 2 (#6949)
• Avoid using ImportError during import cupy (#6954)
• Fix cuDNN descriptor double destroy (#6972)
• Fix ConcatWithAsyncTransfer (#6992)
• Set allow_pickle=True (#7036)
• Fix subview of zero-sized arrays (#7037)
• Fix At output offset (#7046)
• Fix handling of ndarray offsets (#7047)
• Fix construction of std::shared_ptr with custom deleter in chianer_interop.cc (#7107)
• Fix build with clang (#7119)

Code Fixes

• Check headers with clang-tidy (#6441)
• Refactor CUDA batch norm tensor descriptor (#6724)
• Fix comments and add TODO to indexing routines (#6789)
• Add cuda_internal::DeviceInternals to wrap handle etc. (#6820)
• Clean up DeviceInternals (#6827)
• Rename CHAINERX_REGISTER_OP_{NATIVE,CUDA} to CHAINERX_{NATIVE,CUDA}_REGISTER_OP (#6865)
• Add comments on del (#6933)
• Unify variable names in gradient_check (#6935)
• Align macro parameter name (#6941)
• Introduce chainerx/kernels/ and rename existing device "op"s to "kernel"s (#6944)
• Remove obsolete "Op" files (#6959)
• Prefix macro with CHAINERX as per convention (#7022)
• Use macro in exp_log.{cc/cu} (#7068)
• Pass arguments by value in native::Float16 and cuda::Float16 (#7069)
• Avoid importing object (#7110)

Documentation

• Fix to clarify the description about initializer argument (#6317)
• Add docs for two loss functions (#6349, thanks @hsezhiyan!)
• Improve docs of square, maximum and squared_difference (#6451, thanks @aksub99!)
• Append to v6 upgrade guide about Python 3.4 support drop (#6493)
• Add reference and warning to F.swish document (#6509, thanks @fiarabbit!)
• Document fix in default initializer (#6519)
• Convert utilities docs to one page (#6595, thanks @trancenoid!)
• Add chainer.get_device to doc (#6735)
• Use search index (#6881)
• Add chainerx.sigmoid docs (#6889, thanks @crcrpar!)
• Fix typo in F.convolution_2d (#6890, thanks @crcrpar!)
• Document chainer.testing.LinkTestCase (#6895, thanks @crcrpar!)
• Update README.txt for a link to the tutorial (#6896)
• Fix broken link in chainerx.md (#6899, thanks @tkat0!)
• Document passive attributes in FunctionTestCase (#6931)
• Fix documentation of renamed arguments (#6932)
• Fix typo in pickle_dataset.py (#6942)
• Update ChainerX contribution guide (#6951)
• Support Sphinx 2.0 and use absolute path to support the latest RTD (#7027)
• Fix link to ChainerMN docs in performance guide (#7044)
• Update supported MPI list (#7086)
• Document CHAINERX_ENABLE_BLAS environment variable (#7098, thanks @durswd!)
• Move backend docs to a separate page (#7099)
• Document backend and device objects (#7102)
• Remove extra spaces in docstrings (#7125)
• Fix AdamW docstring (#7137, thanks @crcrpar!)
• Fix spelling of AMSGrad (#7138, thanks @crcrpar!)

Installation

• CMake for Windows(clang-cl) (#7039, thanks @durswd!)
• Exclude protobuf 3.8.0rc1 from dependencies (#7083)

Examples

• Improve chainer examples (#6399, thanks @crcrpar!)
• Fix reinforcement_learning example to work with default dtype (#6624)
• Support default dtype in vae example (#6717)
• Support ChainerX in reinforcement learning example (#6733)
• Support ChainerX in wavenet example (#6736)
• Trivial fixes to Wavenet example (#6737)
• Support ChainerX in VAE example (#6739)
• Support ChainerX in text classification example (#6769)
• Support ChainerX in DCGAN example (#6773)
• Support ChainerX in word2vec example (#6774)
• Show download progress bar in image-captioning example (#6775)
• Support ChainerX in memnn example (#6854)
• Use filename in PlotReport example (#6880, thanks @crcrpar!)
• Support ChainerX in CIFAR example (#6936)
• Support ChainerX in POS-tagging example (#7081)
• Support ChainerX in Sentiment example (#7087)
• Add progress bar to sentiment analysis example (#7103)
• Support ChainerX in Model Zoo example (#7129)

Tests

• Simplify F.mean_absolute_error test (#6253, thanks @aksub99!)
• Simplify F.bilinear test (#6488, thanks @ishanrai05!)
• Simplify F.deconvolution_2d test (#6498, thanks @ishanrai05!)
• Display pytest summary (#6625, thanks @kshitij12345!)
• Travis test against v6 branch (#6749)
• Fix Travis with macOS (#6754)
• Dodge nondifferentiable inputs in chainerx.max test (#6761)
• Make too slow initializers' tests faster (#6792)
• Fix test failures in math test (#6798)
• Simplify F.flip test (#6801, thanks @ishanrai05!)
• Simplify F.where test (#6802, thanks @ishanrai05!)
• Simplify F.repeat test (#6803, thanks @ishanrai05!)
• Fix F.elu test numeric error (#6841)
• Relax tolerance for float16 in unary_math_function_unittest (#6845)
• Relax tolerances and avoid non-differentiable points for FP16 in triplet loss tests (#6855)
• Simplify F.unpooling_nd test (#6861, thanks @ishanrai05!)
• Simplify F.local_response_normalization test (#6867, thanks @ishanrai05!)
• Simplify F.reshape test (#6868, thanks @ishanrai05!)
• Simplify F.layer_normalization test (#6871, thanks @ishanrai05!)
• Fix test failure in test_spatial_transformer_sampler.py (#6883)
• Simplify F.prelu test (#6887, thanks @ishanrai05!)
• Simplify F.flatten test (#6888, thanks @ishanrai05!)
• Simplify F.dstack test (#6891, thanks @ishanrai05!)
• Simplify F.sign test (#6898, thanks @hikjik!)
• Simplify F.ceil test (#6900, thanks @hikjik!)
• Simplify F.floor test (#6901, thanks @hikjik!)
• Fix F.rrelu test instability (#6915)
• Fix F.max_pooling_nd test instability (#6917)
• Fix flaky Huber loss test (#6924)
• Simplify F.fmod test (#6937, thanks @hikjik!)
• Simplify F.fix test (#6938, thanks @hikjik!)
• Fix test parameters in ChainerX math tests (#6946)
• Increase the default columns in Travis CI (#6948)
• Fold Travis test outputs (#6961)
• Simplify 'F.min', 'F.max' test (#6962, thanks @hikjik!)
• Simplify 'F.exp', 'F.log' test (#6963, thanks @hikjik!)
• Simplify F.expm1 test (#6965, thanks @hikjik!)
• Fix flaky ChainerX max_pool test (#6975)
• Simplify F.bias test (#6976, thanks @hikjik!)
• Simplify F.cumsum test (#6977, thanks @hikjik!)
• Refactor Variable.addgrad test (#6979)
• Simplify F.cosh, F.sinh test (#6980, thanks @hikjik!)
• Simplify F.log1p test (#6981, thanks @hikjik!)
• Simplify F.linear_interpolate test (#6984, thanks @hikjik!)
• Simplify F.fft, F.ifft test (#6985, thanks @hikjik!)
• Simplify F.matmul test (#6987, thanks @ishanrai05!)
• Fix flaky TestLogSumExp (#6988)
• Fix flaky TestMin (#6989)
• Simplify F.get_item test (#6990)
• Simplify F.inv, F.batch_inv test (#6994, thanks @hikjik!)
• Simplify F.batch_l2_norm_squared test (#6996, thanks @hikjik!)
• Simplify F.accuracy test (#7006, thanks @hikjik!)
• Simplify F.binary_accuracy test (#7007, thanks @hikjik!)
• Simplify F.r2_score test (#7008, thanks @hikjik!)
• Simplify F.permutate test (#7010, thanks @hikjik!)
• Simplify F.scatter_add test (#7012, thanks @hikjik!)
• Simplify F.separate test (#7013, thanks @hikjik!)
• Simplify F.logsumexp test (#7018, thanks @hikjik!)
• Skip tests that fail with NumPy 1.16.3 (#7021)
• Add broadcast test in test_math.py (#7023)
• Fix flaky chainerx.abs test (#7024)
• Remove ChainerX acceptance tests (#7026)
• Fix flaky chainerx.tan test (#7033)
• Display pytest summary (cont.) (#7089)

Others

• Make it easier to copy the instruction in the issue template (#6665)
• Make git ignore chainerx/libchainerx.dylib (#6666)
• Add .mergify.yml (#7074)
• Improve mergify configuration (#7111)