Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix comparison with casting in hdf5 serializer test #8182

Merged
merged 2 commits into from
Sep 30, 2019
Merged

Fix comparison with casting in hdf5 serializer test #8182

merged 2 commits into from
Sep 30, 2019

Conversation

niboshi
Copy link
Member

@niboshi niboshi commented Sep 26, 2019 •
edited
Loading

Fixes #8181

Succeeded 100,000 times without a failure.

@niboshi niboshi added cat:test Test or CI related. to-be-backported Pull request that should be backported. labels Sep 26, 2019
emcastillo
emcastillo approved these changes Sep 30, 2019
View reviewed changes
Copy link
Member

@emcastillo emcastillo left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@emcastillo
Copy link
Member

Jenkins, test this please

@chainer-ci
Copy link
Member

Jenkins CI test (for commit afaf8e2, target branch master) failed with status FAILURE.

@emcastillo
Copy link
Member

Py2 compat error detected.
Should we fix it until jenkins support can be dropped?

@niboshi
Copy link
Member Author

niboshi commented Sep 30, 2019

@emcastillo This needs to be backported. I fixed the error.

@emcastillo
Copy link
Member

Jenkins, test this please

@chainer-ci
Copy link
Member

Jenkins CI test (for commit 76a3b27, target branch master) failed with status FAILURE.

@emcastillo
Copy link
Member

Another related failure 😥

@niboshi
Copy link
Member Author

niboshi commented Sep 30, 2019

six.BytesIO does not work in Python 2 for some reason.
I replaced with tempfile.TemporaryFile and it's fixed.

@emcastillo
Copy link
Member

Jenkins, test this please

@chainer-ci
Copy link
Member

Jenkins CI test (for commit b698d6a, target branch master) succeeded!

@emcastillo emcastillo added the st:test-and-merge State indicating that pull request is approved by a reviewer and can be merged after CI passes. label Sep 30, 2019
@mergify mergify bot merged commit adec60d into chainer:master Sep 30, 2019
emcastillo pushed a commit to emcastillo/chainer that referenced this pull request Sep 30, 2019
@mergify
Merge pull request chainer#8182 from niboshi/test-hdf5
f4e4339 
Fix comparison with casting in hdf5 serializer test
@emcastillo emcastillo mentioned this pull request Sep 30, 2019
Merged
@niboshi niboshi deleted the test-hdf5 branch September 30, 2019 08:22
toslunar added a commit to toslunar/chainer that referenced this pull request Oct 2, 2019
@toslunar
Make chainer#8182 work with h5py<2.9
bc93d87
@toslunar toslunar mentioned this pull request Oct 2, 2019
Merged
Crissman added a commit to Crissman/chainer that referenced this pull request Oct 9, 2019
@Crissman
# This is a combination of 2 commits.
6502ee5 
# The first commit's message is:

# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 5 commits.
# The first commit's message is:
# This is a combination of 24 commits.
# The first commit's message is:
# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 23 commits.
# The first commit's message is:
# This is a combination of 22 commits.
# The first commit's message is:
# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 25 commits.
# The first commit's message is:
# This is a combination of 62 commits.
# The first commit's message is:
# This is a combination of 3 commits.
# The first commit's message is:
# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 5 commits.
# The first commit's message is:
# This is a combination of 109 commits.
# The first commit's message is:
# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 2 commits.
# The first commit's message is:

# This is a combination of 86 commits.
# The first commit's message is:
# This is a combination of 12 commits.
# The first commit's message is:
# This is a combination of 27 commits.
# The first commit's message is:
matrix initialization fix

# This is the 2nd commit message:

Simplify ArrayReprImpl

# This is the 3rd commit message:

NoBackpropModeScope

# This is the 4th commit message:

Insert a newline per 10 items

# This is the 5th commit message:

Abbreviate output of ArrayRepr for large inputs

# This is the 6th commit message:

Add test

# This is the 7th commit message:

Raise FutureWarning on GPU-to-GPU transfer in StandardUpdater

# This is the 8th commit message:

Fix StandradUpdater test to handle FutureWarning on GPU-to-GPU transfer

# This is the 9th commit message:

Fix LSTM omitted upstream gradients

# This is the 10th commit message:

Always get typeid of kernels in libchainerx

Apparently, typeid can be inconsistent among multiple binary
modules on MacOSX (and probably on Windows).

# This is the 11th commit message:

Run clang-format

# This is the 12th commit message:

Update kernels for recent ChainerX

# This is the 13th commit message:

Move kernel key implementations to chainerx_native

# This is the 14th commit message:

Register RNN key kernels

# This is the 15th commit message:

Fix the unittest for KernelRegistry

# This is the 16th commit message:

Allow to define kernel outside chainerx namespace

# This is the 17th commit message:

Fix build

# This is the 18th commit message:

Handle fabs kernel removal

# This is the 19th commit message:

Register recently added kernels

# This is the 20th commit message:

Add a TODO comment for restriction about namespace

# This is the 21st commit message:

Update the link to the interface change

# This is the 22nd commit message:

Support device arrays when initializing `variable.Parameter` objects

# This is the 23rd commit message:

copy device arrays to numpy ones in Parameter initialization

# This is the 24th commit message:

format fixes

# This is the 25th commit message:

fix numpy version in Dockerfile

# This is the 26th commit message:

Add float16 test in array_repr_test.cc

# This is the 27th commit message:

Add a comment

# This is the 2nd commit message:

Avoid non-differentiable point manulally

# This is the 3rd commit message:

Add a comment

# This is the 4th commit message:

Fix -Wabsolute-value

# This is the 5th commit message:

Zero sized inputs for 'solve' are now working

# This is the 6th commit message:

Zero sized inputs for 'svd' are now working

# This is the 7th commit message:

Zero sized inputs for 'pinv' are now working

# This is the 8th commit message:

Zero sized inputs for 'inv' are now supported and tested

# This is the 9th commit message:

Skip tests if zero sized input is not supported by numpy

# This is the 10th commit message:

clang-format

# This is the 11th commit message:

Skip checking strides for inverse tests

# This is the 12th commit message:

Moved Identity for zero sized input to kernels

# This is the 2nd commit message:

Fixed error for zero-sized input for native dot kernel

# This is the 3rd commit message:

clang-format

# This is the 4th commit message:

Moved changes in dot kernels to the separate PR chainer#8007

# This is the 5th commit message:

Use local variable instead of std::max in function call

# This is the 6th commit message:

Fixed unused parameter 'compute_uv' in unavailable LAPACK code path

# This is the 7th commit message:

clang-format

# This is the 8th commit message:

TrueDiv tolerances

# This is the 9th commit message:

Fix doc of `backend.copyto`

# This is the 10th commit message:

Fix native AddAt implementation for float16 arrays

# This is the 11th commit message:

refactoring batch normalization

This commit refactors chainer batch normalization class.
We extract the three backends "iDeep", "cudnn", "general cpu gpu" into
separate plugins, which increases the readabilities and make the
integration with `chainermn.batch_normalization` easier.

# This is the 12th commit message:

update batch normalization and fix bugs

# This is the 13th commit message:

separate the batch_normalization_grad

# This is the 14th commit message:

clean up and fix bugs in batch_normalization

# This is the 15th commit message:

Merge Chainermn batch normalization into Chainer

This commit refactors `batch_normalization` code in following ways:
1. removes all the outdated codes
2. merge the MultiNodeBatchNormalization into chainer.
   - make `MultiNodeBatchNormalization` a derived class to BatchNormalization
   in `chainer`
   - make `_NcclBackend` & `_MpiBackend` derived from
   `_GeneralBatchNormalizationBackend`

   to use the codes in chainer

# This is the 16th commit message:

refactoring link/bn to the new implementation

# This is the 17th commit message:

move communication backend check to __init__

# This is the 18th commit message:

clean up

# This is the 19th commit message:

clean up

# This is the 20th commit message:

fix for pep8

# This is the 21st commit message:

remove abc to align with chainer coding style

# This is the 22nd commit message:

include cuda elementwise optimization from chainermn

# This is the 23rd commit message:

replace mode with necessary fields only

# This is the 24th commit message:

fix bugs

# This is the 25th commit message:

Move backend selector to a callable class

This commit move the batch normalization backend selector from the
member function to a separate callable class, to eliminate the necessity for
chainermn batch normalization to inherit from the chainer one.

# This is the 26th commit message:

remove unused import

# This is the 27th commit message:

Rename _GeneralBatchNormalizationBackend

Remove underscore as it is imported in chainermn

# This is the 28th commit message:

remove unnecessary mode in backward

# This is the 29th commit message:

remove mode from selector parameters

# This is the 30th commit message:

remove underscore

# This is the 31st commit message:

align coding style

# This is the 32nd commit message:

change member variable to return variable

# This is the 33rd commit message:

rename backend to Impl

# This is the 34th commit message:

remove outdated comment

# This is the 35th commit message:

Generalize and simplify native CumsumKernel implementation

# This is the 36th commit message:

add cblas.h and modify CMakeLists.txt

# This is the 37th commit message:

add comment

# This is the 38th commit message:

Treat warnings as errors in tests

# This is the 39th commit message:

Remove obsolete ignore warning entry

# This is the 40th commit message:

Ignore RuntimeWarning chainer#8062

# This is the 41st commit message:

Suppress number of combinations of in_out_dtype

# This is the 42nd commit message:

Fix braced-scalar-init warning

# This is the 43rd commit message:

Add mean_squared_error for chainerX

# This is the 44th commit message:

Fix python code style

# This is the 45th commit message:

Explicit h5py.File mode due to deprecation

# This is the 46th commit message:

Add test for optimizers test coverage

# This is the 47th commit message:

use standard constructor

# This is the 48th commit message:

Apply suggestions from code review

OK, these hide one move...

Co-Authored-By: Hiroyuki Vincent Yamazaki <hiroyuki.vincent.yamazaki@gmail.com>
# This is the 49th commit message:

Document chainerx.to_numpy

# This is the 50th commit message:

Fix expected

# This is the 51st commit message:

Fix input generation

# This is the 52nd commit message:

Fix style

# This is the 53rd commit message:

Avoid computing with uninitialized array

`np.empty(...) * 2` may cause
`RuntimeWarning: overflow encountered in multiply`

# This is the 54th commit message:

Fix fill value different dtype constant initialization

# This is the 55th commit message:

fix 8084

# This is the 56th commit message:

Fix heading anchor in ChainerX docs

# This is the 57th commit message:

Use some latest features for the WaveNet example

# This is the 58th commit message:

Add timestamp to Travis CI log

# This is the 59th commit message:

Avoid unnecessary test condition

# This is the 60th commit message:

prevent zero eigenvectors

# This is the 61st commit message:

Add scripts for Docker base images for Chainer CI.

# This is the 62nd commit message:

Resolve absolute path.

# This is the 63rd commit message:

Revive clang-tidy test in Travis CI

# This is the 64th commit message:

clang-tidy fixes

# This is the 65th commit message:

Chainerx loss_scale support

# This is the 66th commit message:

calc loss scaling after backward pass

# This is the 67th commit message:

fixed comments

# This is the 68th commit message:

Call super on multiprocess_parallel_updater finalize

# This is the 69th commit message:

Fix coding style of some imports in ChainerMN

# This is the 70th commit message:

Support mixed16/float16 GroupNormalization

# This is the 71st commit message:

Remove unnecessary code

# This is the 72nd commit message:

Support parameterized dtype in group normalization tests

# This is the 73rd commit message:

Fix flaky test for group normalization

# This is the 74th commit message:

Remove a harmful dtype argument

# This is the 75th commit message:

Add pytest.ini for custom markers

# This is the 76th commit message:

Add more markers

# This is the 77th commit message:

Add chainerx marker

# This is the 78th commit message:

Fix test failure warning checking

# This is the 79th commit message:

Revert other fix

# This is the 80th commit message:

Move pytest markers to `setup.cfg`

# This is the 81st commit message:

Remove duplicated parameter

# This is the 82nd commit message:

Use pytest-style parameter destruction

See chainer#6635

# This is the 83rd commit message:

Fix test class name

# This is the 84th commit message:

Fix Travis homebrew update

# This is the 85th commit message:

Update abseil to new release

# This is the 86th commit message:

Remove TODO comment

# This is the 2nd commit message:

Remove unused include

# This is the 2nd commit message:

Register builtin key kernel

# This is the 2nd commit message:

Add CUDA ScanKernel

# This is the 3rd commit message:

Implement CudaCumsumKernel

# This is the 4th commit message:

Cumsum cuda test

# This is the 5th commit message:

Add tests for large array case

# This is the 6th commit message:

Fix variable name for the readability

# This is the 7th commit message:

Fix variable name to reduce_len

# This is the 8th commit message:

To avoid an error on DALI, making a new module for dataset.

# This is the 9th commit message:

Fix a few lines to follow the style guide by flake8.

# This is the 10th commit message:

Remove unused headers

# This is the 11th commit message:

Remove TODO fixed by chainer#7528

# This is the 12th commit message:

Make parameterized test names deterministic

# This is the 13th commit message:

Revert "Add mean_squared_error for chainerX"

# This is the 14th commit message:

Relax tolerance of im2col test

# This is the 15th commit message:

Add native implementation

# This is the 16th commit message:

Modify implementation to use elementwise kernel

# This is the 17th commit message:

Add tests

# This is the 18th commit message:

Remove unnecessary kernel implementation

# This is the 19th commit message:

Split output for multidimensional inputs

# This is the 20th commit message:

Workaround for problem with repeated indices

# This is the 21st commit message:

Fix implementation for multidimensional cases

# This is the 22nd commit message:

Add 0 dim case

# This is the 23rd commit message:

Remove unused header

# This is the 24th commit message:

Fix segmentation error

# This is the 25th commit message:

Fix test errors

# This is the 26th commit message:

Remove 0-dim cases

# This is the 27th commit message:

Modify return type to tuple

# This is the 28th commit message:

modify routines.cc

# This is the 29th commit message:

Make chainerx.nonzero return tuple

# This is the 30th commit message:

Fix bug

# This is the 31st commit message:

Fix tests

# This is the 32nd commit message:

Add docs

# This is the 33rd commit message:

Filter DeprecationWarning in test_maniplation.py

# This is the 34th commit message:

Use warnings.catch_warnings

# This is the 35th commit message:

Allow fix_random decorator to be used with OpTest

# This is the 36th commit message:

Fix flaky RNN tests by fixing random seed

# This is the 37th commit message:

Introduce op_utils.fix_random

# This is the 38th commit message:

Remove unrelated change for PR

# This is the 39th commit message:

Support all float dtyeps in F.n_step_rnn and F.n_step_birnn

# This is the 40th commit message:

Adjust forward test tolerance

# This is the 41st commit message:

Use narrower input ranges for FP16 test

# This is the 42nd commit message:

Implemented zeta function

# This is the 43rd commit message:

Added test for zeta

# This is the 44th commit message:

Added documentation for zeta

# This is the 45th commit message:

Confirmed backward code

# This is the 46th commit message:

Changes for zeta_test

# This is the 47th commit message:

fixed flake8 issue

# This is the 48th commit message:

modified backward and added few tests

# This is the 49th commit message:

fixed bugs

# This is the 50th commit message:

modified backward implementation

# This is the 51st commit message:

adjusted tolerances

# This is the 52nd commit message:

accomodated changes

# This is the 53rd commit message:

accomodated changes

# This is the 54th commit message:

fixed dtype

# This is the 55th commit message:

Added changes

# This is the 56th commit message:

Update chainer/functions/math/zeta.py

added changes

Co-Authored-By: Hiroyuki Vincent Yamazaki <hiroyuki.vincent.yamazaki@gmail.com>
# This is the 57th commit message:

Update tests/chainer_tests/functions_tests/math_tests/test_zeta.py

Co-Authored-By: Hiroyuki Vincent Yamazaki <hiroyuki.vincent.yamazaki@gmail.com>
# This is the 58th commit message:

Changed documentation

# This is the 59th commit message:

Modified exception shape and testing shape

# This is the 60th commit message:

removed parametrization

# This is the 61st commit message:

Added shape for parametrization

# This is the 62nd commit message:

Add ignore::ImportWarning to setup.cfg

# This is the 63rd commit message:

Ignore specific ImportWarning, add comments

# This is the 64th commit message:

Remove obsolete note in `chainerx.maximum` doc

# This is the 65th commit message:

Fixes deepcopy for chain parameters

# This is the 66th commit message:

fixed incomplete tests

# This is the 67th commit message:

fixed params initial value

# This is the 68th commit message:

[chainerx::vsplit] Implements vsplit in chainerx

# This is the 69th commit message:

[chainerx::vsplit] Adds Python bindings

# This is the 70th commit message:

[chainerx::vslpit] Adds documentations

# This is the 71st commit message:

[chainerx::vsplit] Adds tests

# This is the 72nd commit message:

Add documentation for `chainerx.minimum`

# This is the 73rd commit message:

Add `testing.generate_matrix`

# This is the 74th commit message:

Remove condiiton_numbers for simplicity

# This is the 75th commit message:

FIx minor bugs and styles

# This is the 76th commit message:

Add tests of generate_matrix

# This is the 77th commit message:

Apply suggestions from code review

Co-Authored-By: Toshiki Kataoka <tos.lunar@gmail.com>
# This is the 78th commit message:

Improve docstring, remove unused variable

# This is the 79th commit message:

Add tests and fix for broadcast in generate_matrix

# This is the 80th commit message:

Remove unused import

# This is the 81st commit message:

Improve error types and messages

# This is the 82nd commit message:

Support NumPy 1.9 in generate_matrix

# This is the 83rd commit message:

Use uniform instead of random+add const

Co-Authored-By: Toshiki Kataoka <tos.lunar@gmail.com>
# This is the 84th commit message:

Support numpy1.9 in generate_matrix tests

# This is the 85th commit message:

Fix flaky RNN tests by fixing random seed

# This is the 86th commit message:

Avoid NOQA

# This is the 87th commit message:

Document CHAINERX_ENABLE_LAPACK

# This is the 88th commit message:

Improve documentation of ChainerX environment variables

# This is the 89th commit message:

Make explanations more accurate

# This is the 90th commit message:

Update docs/source/chainerx/install/index.rst

Co-Authored-By: Crissman Loomis <crissman@preferred.jp>
# This is the 91st commit message:

Update docs/source/chainerx/install/index.rst

Co-Authored-By: Crissman Loomis <crissman@preferred.jp>
# This is the 92nd commit message:

Convert tabs to spaces in setup.cfg

# This is the 93rd commit message:

Fix missing dtype checks in test_loss

# This is the 94th commit message:

Add mixed dtypes tests

# This is the 95th commit message:

Fix RNN link documentation

# This is the 96th commit message:

Avoid empty in F.cast test that can cause overflow warning

# This is the 97th commit message:

Fix typo

# This is the 98th commit message:

Make xdist usable in ChainerX tests

# This is the 99th commit message:

Make get_runtime_info private method

# This is the 100th commit message:

Add docstring for print_runtime_info

# This is the 101st commit message:

Fix english

# This is the 102nd commit message:

Fix markups

# This is the 103rd commit message:

Remove unnecessary print

# This is the 104th commit message:

Put link to print_runtime_info on backends.rst

# This is the 105th commit message:

Move reference to chainer.print_runtime_info

# This is the 106th commit message:

Update docstring

# This is the 107th commit message:

Update listing style

# This is the 108th commit message:

Adjust tolerances

# This is the 109th commit message:

Bump version numbers

# This is the 2nd commit message:

Fix tests

# This is the 3rd commit message:

Fix bugs in cupy elementwise in batch_normalization

This commit fixes a bug in cupy
elementwise of batch_normalization by assign different dtypes to mean
and var

# This is the 4th commit message:

Fix style of `setup.cfg`

# This is the 5th commit message:

remove dummy serializers

# This is the 2nd commit message:

Add hinge test

# This is the 2nd commit message:

Add norm test parameter

# This is the 3rd commit message:

Remove unused includes

# This is the 2nd commit message:

Make Weight Initializers have its own RandomState instance

# This is the 3rd commit message:

Make the initializers work properly even without CUDA

# This is the 4th commit message:

Receive RandomState instead of seed as the argument and use backend.copyto

# This is the 5th commit message:

Remove unused imports and variables

# This is the 6th commit message:

Use chainer.utils.argument.parse_kwargs

# This is the 7th commit message:

Make the initializers not have None as the rng attribute

# This is the 8th commit message:

Keep backward compatibility

# This is the 9th commit message:

Avoid redundant copy

# This is the 10th commit message:

Remove unnecessary colons in docstrings

# This is the 11th commit message:

Add test parameter 'rng'

# This is the 12th commit message:

Drop tests with cupy.random.RandomState

# This is the 13th commit message:

Avoid sharing rng among tests

# This is the 14th commit message:

Support zero-sized shape in generate_matrix

# This is the 15th commit message:

Raise error on shape mismatch with zero-sized arrays

# This is the 16th commit message:

Fix chainer#8156

`Convolution1D` caused an error because the input shape was still 4D. So converted.

# This is the 17th commit message:

Fix chainer#8156

`generate.py` cause an error the same as `train.py`.

# This is the 18th commit message:

Revive `_gen_case` approach

It newly fixes `__str__` of parameterized class by partially reverting
the commit 108fa46 that caused the
binding issues, two of which are fixed later by
630a2da and chainer#6287.

# This is the 19th commit message:

Fix indent

# This is the 20th commit message:

Fix overflow warning in div backward test

# This is the 21st commit message:

disable flaky test retry in flexCI

# This is the 22nd commit message:

move test_elu

# This is the 23rd commit message:

dedup

# This is the 24th commit message:

Randomly reduce testing parameters

# This is the 25th commit message:

pairwise test

# This is the 26th commit message:

Add comment

# This is the 27th commit message:

Use more index-accesses for readability

# This is the 28th commit message:

refactor

# This is the 29th commit message:

pythonic typing

# This is the 30th commit message:

`test_accuracy.py` is not so slow

# This is the 31st commit message:

Move pairwise impl and patch `testing.product_dict`

# This is the 32nd commit message:

Pairwise testing for ChainerX

# This is the 33rd commit message:

Fix flaky TestQR

# This is the 34th commit message:

Relax BN backward test tolerances

# This is the 35th commit message:

copy to __deepcopy__

Change from overriding `copy` to `__deepcopy__` as deepcopy is used in
link `chainer/link.py:435`

# This is the 36th commit message:

Change `to_gpu` to `to_device`

This commit changes deprecated `to_gpu` to the new `to_device` in the tests

# This is the 37th commit message:

Avoid direct import of cupy

# This is the 38th commit message:

add missing copy

# This is the 39th commit message:

get device only when use gpu

# This is the 40th commit message:

use chainer.backends.cuda

# This is the 41st commit message:

Fix comparison with casting in hdf5 serializer test

# This is the 42nd commit message:

Use tempfile instead of BytesIO

# This is the 43rd commit message:

Allow pytest.skip() in combination with testing.repeat/retry

# This is the 44th commit message:

Fix the way importing _pytest.outcomes

# This is the 45th commit message:

Fix passing args and kwargs

# This is the 46th commit message:

Remove ImportWarning ignore entry

# This is the 47th commit message:

Fix bugs when x and gamma/beta have different dtypes

This commit fixes various bugs in the batch normalization, where the x
and gamma/beta have different dtypes.

# This is the 48th commit message:

Add tests for x and gamma/beta with different dtypes

This commit adds tests for the cases where x and gamma/beta have
different dtypes, which was not tested, but is supposed to be supported.

This commit adds float16 alone with float64 tests for x and gamma/beta,
while for the ideep case, since it only supports float32, hence all the
variables are casted to float32 in ideep tests.

# This is the 49th commit message:

support only float32 for ideep

# This is the 50th commit message:

remove the constrain of float32 in ideep

change the condition to x.dtype == gamma.dtype

# This is the 51st commit message:

break line with ()

# This is the 52nd commit message:

move the astype to impl backward

# This is the 53rd commit message:

support ChainerMN testing in pfnci

# This is the 54th commit message:

preinstall cupy in Docker images, support other versions of CUDA

# This is the 55th commit message:

address review comments

# This is the 56th commit message:

change the structure of ChainerMN tests on pfnci

# This is the 57th commit message:

unsupport Python 2 in pfnci testing for ChainerMN

# This is the 58th commit message:

do not install Python 2 in the Docker image for ChainerMN testing, do not uninstall existing CuPy in test (not installed in the image)

# This is the 59th commit message:

Make CUDA base image configurable in chainermn-ci-prep.sh

# This is the 60th commit message:

fix comment

# This is the 61st commit message:

Add multi-node efficient snapshot

# This is the 62nd commit message:

Remove unnecessary default argument

# This is the 2nd commit message:

New configuration system of communicator inspired by links

# This is the 3rd commit message:

Enable batched_copy by default on AllReduce

# This is the 4th commit message:

Add test on config scope

# This is the 5th commit message:

Fix to work with allreduce grad type None

# This is the 6th commit message:

Fix test code

# This is the 7th commit message:

add bilinear initializer

# This is the 8th commit message:

fix typo in test_bilinear

# This is the 9th commit message:

avoid redefining filter

# This is the 10th commit message:

Generalize Bilinear for ND input

# This is the 11th commit message:

rename initializer: Bilinear -> UpsamplingDeconvFilter

# This is the 12th commit message:

rename _get_upsampling_filter -> _get_bilinear_upsampling_filter

# This is the 13th commit message:

add DownsamplingConvFilter for downsampling

# This is the 14th commit message:

add DownsamplingConvFilter doc
# This is the 15th commit message:

check sum of upsampling/downsampling initializer output

# This is the 16th commit message:

update sampling initializer docstring

# This is the 17th commit message:

rename to SamplingFilter

# This is the 18th commit message:

bilinear -> liniear

# This is the 19th commit message:

refactor sampling.py

# This is the 20th commit message:

raise ValueError when ksize is not same

# This is the 21st commit message:

refactor test_sampling.py

# This is the 22nd commit message:

test for various ksize

# This is the 23rd commit message:

add stride expectation in documentation

# This is the 24th commit message:

use xp.arange instead of range

# This is the 25th commit message:

use cuda.to_cpu for assertion

# This is the 2nd commit message:

Added cusolver to CMakeLists

# This is the 2nd commit message:

Revert changes in chainerx_cc/chainerx/cuda/cuda_device.cc

(cherry picked from commit 1a6031a)

# This is the 2nd commit message:

Remove changes from chainerx_cc/chainerx/routines/creation.cc

# This is the 2nd commit message:

Decreased eps parameter to increase accuracy of finite difference derivatives

# This is the 2nd commit message:

Use generate_matrix to fix flaky solve tests

# This is the 3rd commit message:

add WIN32_LEAN_AND_MEAN def

Potentially save developer from preprocessing error due to macro expansion when accidentally include `windows.h`, e.g. `nvToolsExt.h` will do this.
# This is the 4th commit message:

Fix a typo

# This is the 5th commit message:

Fix attrs version

# This is the 6th commit message:

Add virtual destructor in CUDA Allocator

# This is the 7th commit message:

Explicitly define special member functions in Allocator

# This is the 8th commit message:

Remove unused classes MaxPoolForwardBackward and AveragePoolForwardBackward

# This is the 9th commit message:

Some code refactoring for ChainerX setuptools

# This is the 10th commit message:

Delete move and copy of virtual classes GradState and ForwardBackward

# This is the 11th commit message:

Reference instead of pointer to avoid copy

# This is the 12th commit message:

fix protobuf dependency

# This is the 13th commit message:

Fix caffe test with protobuf>=3.8

It fixes the following error.
TypeError: Parameter to MergeFrom() must be instance of same class: expected caffe.LayerParameter got dict.

# This is the 14th commit message:

Update resize_images.py
# This is the 15th commit message:

Fix flaky test parmeterization

# This is the 16th commit message:

Make chainer#8182 work with h5py<2.9

# This is the 17th commit message:

Comment why `tempfile.mkstemp`

# This is the 18th commit message:

Set up documentation for loss functions in ChainerX

# This is the 19th commit message:

Relax tolerances in ChainerX unary math tests

# This is the 20th commit message:

Add documentation for chainerx.ravel

# This is the 21st commit message:

Fix possible data race in CUDA memory keeper

# This is the 22nd commit message:

Suppress warnings-as-errors from external libraries

# This is the 2nd commit message:

readability-avoid-const-params-in-decls

# This is the 3rd commit message:

Write CMAKE_GENERATOR when building

# This is the 4th commit message:

Remove unused forward declaration of enum

# This is the 5th commit message:

Use clang-format-6.0 if possible and track the version of clang-format

# This is the 6th commit message:

Update chainerx_cc/scripts/run-clang-format.sh

Co-Authored-By: Yuta Okamoto <okapies@gmail.com>
# This is the 7th commit message:

Update chainerx_cc/scripts/run-clang-format.sh

Co-Authored-By: Yuta Okamoto <okapies@gmail.com>
# This is the 8th commit message:

Small fix

# This is the 9th commit message:

Improve example tests

# This is the 10th commit message:

Follow changes in master

# This is the 11th commit message:

Restore --test option in train_imagenet.py

# This is the 12th commit message:

Add "jenkins" extras

# This is the 13th commit message:

Remove legacy DeprecationWarning filter

This commit removes the lagacy filters.

# This is the 14th commit message:

Add fp16 test to n_step_rnn tests

This commit adds the fp16 tests to the n_step_rnn tests.
The chainer.testing is used to parameterize and switch between the
float16 and float32.

# This is the 15th commit message:

Change to use pytest.mark.parametrize

related to issue chainer#7757

# This is the 16th commit message:

remove legacy filter

# This is the 17th commit message:

remove remainder filter

# This is the 18th commit message:

Add libchainerx_base.a to link chainerx statically

# This is the 19th commit message:

Add an empty file to make old CMake happy

# This is the 20th commit message:

Update chainerx_cc/chainerx/empty.cc

Co-Authored-By: niboshi <niboshi000@gmail.com>
# This is the 21st commit message:

Add documentation for chainerx.sigmoid_cross_entropy

# This is the 22nd commit message:

Flake8

# This is the 23rd commit message:

Fix argument names in documentation

# This is the 2nd commit message:

add ValueError

# This is the 2nd commit message:

address comments

* remove unused ValueError.
* remove unused <string> import.

# This is the 2nd commit message:

move onnx-chainer modules

# This is the 3rd commit message:

add test project for onnx-chainer

# This is the 4th commit message:

fix current installer and doc config

# This is the 5th commit message:

apply latest flake8

# This is the 6th commit message:

add modules for doctest

# This is the 7th commit message:

fix pydoc on example code

# This is the 8th commit message:

Fix example codes to pass doctest

# This is the 9th commit message:

fix code block

# This is the 10th commit message:

fix test script

# This is the 11th commit message:

merge PR244(onnx-chainer)

# This is the 12th commit message:

Deprecate multinode checkpointer

# This is the 13th commit message:

Add one more DeprecationWarning filter

# This is the 14th commit message:

Remove example for deprecated checkpointer

# This is the 15th commit message:

Add missing warnings

# This is the 16th commit message:

Fix import order and grouping

# This is the 17th commit message:

[chainerx::hsplit] Implements hsplit in chainerx

# This is the 18th commit message:

[chainerx::hsplit] Adds Python bindings

# This is the 19th commit message:

[chainerx::hsplit] Adds documentation

# This is the 20th commit message:

[chainerx::hsplit] Adds tests

# This is the 21st commit message:

Refactor ideep workarounds in F.batch_normalization

# This is the 22nd commit message:

Remove obsolete comment

# This is the 23rd commit message:

WIP

# This is the 24th commit message:

basic chainerx integration

# This is the 2nd commit message:

revert unnecessary change

# This is the 3rd commit message:

add comment

# This is the 4th commit message:

minor fix

# This is the 5th commit message:

fix pep8

# This is the 2nd commit message:

fix pep8

# This is the 2nd commit message:

add comments
@kmaehashi kmaehashi added this to the v7.0.0rc1 milestone Oct 25, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@emcastillo emcastillo emcastillo approved these changes

Assignees

@emcastillo emcastillo

Labels
cat:test Test or CI related. st:test-and-merge State indicating that pull request is approved by a reviewer and can be merged after CI passes. to-be-backported Pull request that should be backported.
Projects
None yet
Milestone
v7.0.0rc1
4 participants
@niboshi @emcastillo @chainer-ci @kmaehashi