[incubator-mxnet] branch master updated (90091b1 -> 5c136c9)
This is an automated email from the ASF dual-hosted git repository. patriczhao pushed a change to branch master in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git. from 90091b1 [Numpy] Numpy copysign (#15851) add 5c136c9 Update dmlc-core (#16149) No new revisions were added by this update. Summary of changes: 3rdparty/dmlc-core| 2 +- amalgamation/amalgamation.py | 3 ++- .../macros/src/main/scala/org/apache/mxnet/utils/CToScalaUtils.scala | 4 ++-- 3 files changed, 5 insertions(+), 4 deletions(-)
[GitHub] [incubator-mxnet] pengzhao-intel merged pull request #16149: update dmlc-core
pengzhao-intel merged pull request #16149: update dmlc-core URL: https://github.com/apache/incubator-mxnet/pull/16149 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ZhennanQin commented on issue #16149: update dmlc-core
ZhennanQin commented on issue #16149: update dmlc-core URL: https://github.com/apache/incubator-mxnet/pull/16149#issuecomment-531649192 @szha Finally, this got CI passed. Please review and merge. Thanks a lot. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] nacorti commented on issue #15320: Weird C++ Error / Bug when calling asnumpy() or exporting the weight of darknet53 while training
nacorti commented on issue #15320: Weird C++ Error / Bug when calling asnumpy() or exporting the weight of darknet53 while training URL: https://github.com/apache/incubator-mxnet/issues/15320#issuecomment-531645128 Same problem here, also in python3 Using a custom dataset on a pretrained yolo3_darknet53_voc model from gluon model_zoo The value b in the error seems directly proportional to the batch size and inversely related to the number of workers, but I don't know what to change to make the 3549 value larger. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 edited a comment on issue #16016: [numpy] operator ravel, derive from reshape
tingying2020 edited a comment on issue #16016: [numpy] operator ravel, derive from reshape URL: https://github.com/apache/incubator-mxnet/pull/16016#issuecomment-531645036 > @tingying2020 Would you rebase the code? I will merge after that. Rebased. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 commented on issue #16016: [numpy] operator ravel, derive from reshape
tingying2020 commented on issue #16016: [numpy] operator ravel, derive from reshape URL: https://github.com/apache/incubator-mxnet/pull/16016#issuecomment-531645036 > @tingying2020 Would you rebase the code? I will merge after that. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 closed pull request #16016: [numpy] operator ravel, derive from reshape
tingying2020 closed pull request #16016: [numpy] operator ravel, derive from reshape URL: https://github.com/apache/incubator-mxnet/pull/16016 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 opened a new pull request #16016: [numpy] operator ravel, derive from reshape
tingying2020 opened a new pull request #16016: [numpy] operator ravel, derive from reshape URL: https://github.com/apache/incubator-mxnet/pull/16016 Numpy operator ravel which is the same as reshape(x, -1). @haojin2 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 commented on a change in pull request #16124: [numpy] [tvm] operator true_divide
tingying2020 commented on a change in pull request #16124: [numpy] [tvm]
operator true_divide
URL: https://github.com/apache/incubator-mxnet/pull/16124#discussion_r324511648
##
File path: contrib/tvmop/core/umath.py
##
@@ -0,0 +1,112 @@
+ # Licensed to the Apache Software Foundation (ASF) under one
+ # or more contributor license agreements. See the NOTICE file
+ # distributed with this work for additional information
+ # regarding copyright ownership. The ASF licenses this file
+ # to you under the Apache License, Version 2.0 (the
+ # "License"); you may not use this file except in compliance
+ # with the License. You may obtain a copy of the License at
+ #
+ # http://www.apache.org/licenses/LICENSE-2.0
+ #
+ # Unless required by applicable law or agreed to in writing,
+ # software distributed under the License is distributed on an
+ # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ # KIND, either express or implied. See the License for the
+ # specific language governing permissions and limitations
+ # under the License.
+import tvm
+from .. import defop, AllTypes, AllTypesButHalf
+
+def compute_true_divide(dtype, ndim):
+A = tvm.placeholder([tvm.var() for _ in range(ndim)], name='A',
dtype=dtype)
+B = tvm.placeholder([tvm.var() for _ in range(ndim)], name='B',
dtype=dtype)
+if dtype in ["float16", "float32", "float64"]:
+C = tvm.compute([tvm.var() for _ in range(ndim)],
+lambda *index: A[index] / B[index], name='C')
+else:
+C = tvm.compute([tvm.var() for _ in range(ndim)],
+lambda *index: A[index].astype("float64") /
Review comment:
Changed to `float32` now.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 commented on a change in pull request #16124: [numpy] [tvm] operator true_divide
tingying2020 commented on a change in pull request #16124: [numpy] [tvm]
operator true_divide
URL: https://github.com/apache/incubator-mxnet/pull/16124#discussion_r324511593
##
File path: src/operator/numpy/np_true_divide.cc
##
@@ -41,19 +46,97 @@ bool TrueDivideType(const nnvm::NodeAttrs& attrs,
const int lhs_dtype = in_attrs->at(0);
const int rhs_dtype = in_attrs->at(1);
CHECK_EQ(lhs_dtype, rhs_dtype)
-<< "_true_divide currently only supports same dtype for dividend and
divisor";
+ << "_true_divide currently only supports same dtype for dividend and
divisor";
}
- auto is_float = [](const int dtype) {
-return dtype == mshadow::kFloat32 || dtype == mshadow::kFloat64 || dtype
== mshadow::kFloat16;
- };
-
- for (const int dtype : *in_attrs) {
-CHECK(is_float(dtype)) << "_true_divide currently only supports float
dtype";
+ if (IsIntType(in_attrs->at(0))) {
+TYPE_ASSIGN_CHECK(*out_attrs, 0, mshadow::kFloat64);
Review comment:
Changed to `float32`.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] eric-haibin-lin commented on a change in pull request #16175: [Dataset] add shard API
eric-haibin-lin commented on a change in pull request #16175: [Dataset] add shard API URL: https://github.com/apache/incubator-mxnet/pull/16175#discussion_r324509765 ## File path: python/mxnet/gluon/data/dataset.py ## @@ -84,14 +84,13 @@ def shard(self, num_shards, index): """ assert index < num_shards, 'Shard index of out bound: %d out of %d'%(index, num_shards) assert num_shards > 0, 'Number of shards must be greater than 0' +assert index >= 0, 'Index must be non-negative' length = len(self) -shard_len = length // num_shards +shard_len = (length + num_shards - 1) // num_shards # Compute the start index for this partition start = shard_len * index # Compute the end index for this partition -end = start + shard_len -if index == num_shards - 1: -end = length +end = start + shard_len if index < num_shards - 1 else length Review comment: Got it. Thanks a lot to point this out! I'll fix it This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] wkcn commented on a change in pull request #16175: [Dataset] add shard API
wkcn commented on a change in pull request #16175: [Dataset] add shard API URL: https://github.com/apache/incubator-mxnet/pull/16175#discussion_r324503936 ## File path: python/mxnet/gluon/data/dataset.py ## @@ -84,14 +84,13 @@ def shard(self, num_shards, index): """ assert index < num_shards, 'Shard index of out bound: %d out of %d'%(index, num_shards) assert num_shards > 0, 'Number of shards must be greater than 0' +assert index >= 0, 'Index must be non-negative' length = len(self) -shard_len = length // num_shards +shard_len = (length + num_shards - 1) // num_shards # Compute the start index for this partition start = shard_len * index # Compute the end index for this partition -end = start + shard_len -if index == num_shards - 1: -end = length +end = start + shard_len if index < num_shards - 1 else length Review comment: When `length=6, num_shareds=4`, `shared_len = (6 + 4 - 1) // 4 = 2` The interval of the four partitions are `[0, 2), [2, 4), [4, 6), [6, 6)` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. anirudh2290 pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new 8af2e7a Bump the publish timestamp. 8af2e7a is described below commit 8af2e7af8523bd1c228405b59e4f91455fb66ef1 Author: mxnet-ci AuthorDate: Mon Sep 16 01:37:56 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..2f14945 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Mon Sep 16 01:37:56 UTC 2019
[GitHub] [incubator-mxnet] ZhennanQin commented on a change in pull request #16131: Fix for duplicate subgraph inputs/outputs
ZhennanQin commented on a change in pull request #16131: Fix for duplicate
subgraph inputs/outputs
URL: https://github.com/apache/incubator-mxnet/pull/16131#discussion_r324489273
##
File path: src/operator/subgraph/subgraph_property.h
##
@@ -296,8 +296,20 @@ class SubgraphProperty {
*/
virtual void ConnectSubgraphOutputs(const nnvm::NodePtr subgraph_node,
std::vector*
output_entries) const {
+// Collapse output_entries pointing to same NodeEntry
Review comment:
You're right. But changing inside ConnectSubgraphOutputs() won't help for
the backends that override this function, and because of duplicated output
entries, most of backends had already overrided this function. Your proposal
doesn't work for them. Even for newly added backends, they will most likely
continue overriding ConnectSubgraphOutputs() because the outputs connection may
have a different sequence.
To make things more clear as you requested, I suggest to introduce an new
api, maybe called `ConnectSubgraphUniqueOutputs`, then the code would be like,
```
virtual void ConnectSubgraphUniqueOutputs(const nnvm::NodePtr subgraph_node,
std::vector*
output_entries) const {
size_t idx = 0;
for (size_t i = 0; i < output_entries->size(); ++i) {
*output_entries->at(i) = nnvm::NodeEntry{subgraph_node, idx++, 0};
}
virtual void ConnectSubgraphOutputs(const nnvm::NodePtr subgraph_node,
std::vector*
output_entries) const {
std::vector unique_output_entires; // Collect the unique
entry, let's collect the first entry
std::unordered_map>
output_entires_map; // Collect all duplicated
// Collect above infomation
for (auto entry_ptr : output_entries) {
if (*entry_ptr is unique) { // replace this with nnvm::NodeEntryEqual logic
unique_output_entires.push_back(entry_ptr);
} else {
output_entires_map[first_entry_ptr_of_it].push_back(entry_ptr);
}
}
// Pass unique_output_entires to ConnectSubgraphUniqueOutputs and collect
the change
ConnectSubgraphUniqueOutputs(n, _output_entires);
// Now we know how to make the change to all duplicated
for (auto entry_ptr : unique_output_entires) {
auto duplicated_entries = output_entires_map[entry_ptr];
for (auto& i : duplicated_entries) {
*i = *entry_ptr;
}
}
```
Then backend developer can only override ConnectSubgraphUniqueOutputs() to
avoid handle duplicated outputs.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] wkcn commented on issue #16114: improve dataloader signals and messages
wkcn commented on issue #16114: improve dataloader signals and messages URL: https://github.com/apache/incubator-mxnet/pull/16114#issuecomment-531610885 I have a suggestion: Dataloader does not terminate the program but print a warning when timeout. Users decide whether to terminate it. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] wkcn commented on a change in pull request #16175: [Dataset] add shard API
wkcn commented on a change in pull request #16175: [Dataset] add shard API URL: https://github.com/apache/incubator-mxnet/pull/16175#discussion_r324485736 ## File path: python/mxnet/gluon/data/dataset.py ## @@ -64,6 +64,37 @@ def filter(self, fn): from . import FilterSampler return _SampledDataset(self, FilterSampler(fn, self)) +def shard(self, num_shards, index): +"""Returns a new dataset includes only 1/num_shards of this dataset. + +For distributed training, be sure to shard before you randomize the dataset +(such as shuffle), if you want each worker to reach a unique subset. + +Parameters +-- +num_shards : int +A integer representing the number of data shards. +index : int +A integer representing the index of the current shard. + +Returns +--- +Dataset +The result dataset. +""" Review comment: We also need the assertation `index >= 0` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] wkcn commented on a change in pull request #16175: [Dataset] add shard API
wkcn commented on a change in pull request #16175: [Dataset] add shard API URL: https://github.com/apache/incubator-mxnet/pull/16175#discussion_r324486007 ## File path: python/mxnet/gluon/data/dataset.py ## @@ -64,6 +64,37 @@ def filter(self, fn): from . import FilterSampler return _SampledDataset(self, FilterSampler(fn, self)) +def shard(self, num_shards, index): +"""Returns a new dataset includes only 1/num_shards of this dataset. + +For distributed training, be sure to shard before you randomize the dataset +(such as shuffle), if you want each worker to reach a unique subset. + +Parameters +-- +num_shards : int +A integer representing the number of data shards. +index : int +A integer representing the index of the current shard. + +Returns +--- +Dataset +The result dataset. +""" +assert index < num_shards, 'Shard index of out bound: %d out of %d'%(index, num_shards) +assert num_shards > 0, 'Number of shards must be greater than 0' +length = len(self) +shard_len = length // num_shards +# Compute the start index for this partition +start = shard_len * index +# Compute the end index for this partition +end = start + shard_len +if index == num_shards - 1: +end = length Review comment: The sampler is not uniform. If there are 199 samples and 100 partitions, there will be `1+99` samples in the last partition, but `1` sample in other partitions. The following implementation will be more uniform. ```python shared_len = length // num_shareds rest = length % num_shareds start = shared_len * index + min(index, rest) end = start + shared_len + (index < rest) ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. anirudh2290 pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new 81b2a9c Bump the publish timestamp. 81b2a9c is described below commit 81b2a9c5756bdbb5a7c4e9616d4568c38ebb0bad Author: mxnet-ci AuthorDate: Sun Sep 15 21:14:04 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..0911cea --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Sun Sep 15 21:14:04 UTC 2019
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. anirudh2290 pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new 4ee43b4 Bump the publish timestamp. 4ee43b4 is described below commit 4ee43b4ccc5c653529689b845edc65478bcf6b98 Author: mxnet-ci AuthorDate: Sun Sep 15 19:42:33 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..0d31704 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Sun Sep 15 19:42:33 UTC 2019
[GitHub] [incubator-mxnet] sxjscience edited a comment on issue #16016: [numpy] operator ravel, derive from reshape
sxjscience edited a comment on issue #16016: [numpy] operator ravel, derive from reshape URL: https://github.com/apache/incubator-mxnet/pull/16016#issuecomment-531588040 @tingying2020 Would you rebase the code? I will merge after that. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[incubator-mxnet] branch master updated: [Numpy] Numpy copysign (#15851)
This is an automated email from the ASF dual-hosted git repository.
sxjscience pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
The following commit(s) were added to refs/heads/master by this push:
new 90091b1 [Numpy] Numpy copysign (#15851)
90091b1 is described below
commit 90091b155d6f53c070e3c406f9edc69f38d02e96
Author: Haozheng Fan
AuthorDate: Mon Sep 16 02:57:51 2019 +0800
[Numpy] Numpy copysign (#15851)
* add numpy compatible copysign
* fix scalar op registration error
* add test
---
python/mxnet/ndarray/numpy/_op.py | 53 -
python/mxnet/numpy/multiarray.py | 53 -
python/mxnet/symbol/numpy/_symbol.py | 36 -
src/operator/mshadow_op.h | 10 +++
src/operator/numpy/np_elemwise_broadcast_op.cc | 36 +
src/operator/numpy/np_elemwise_broadcast_op.cu | 21 +
src/operator/operator_tune.cc | 5 ++
tests/python/unittest/test_numpy_op.py | 105 +
8 files changed, 316 insertions(+), 3 deletions(-)
diff --git a/python/mxnet/ndarray/numpy/_op.py
b/python/mxnet/ndarray/numpy/_op.py
index 671345c..b8e4f3f 100644
--- a/python/mxnet/ndarray/numpy/_op.py
+++ b/python/mxnet/ndarray/numpy/_op.py
@@ -33,7 +33,7 @@ __all__ = ['zeros', 'ones', 'full', 'add', 'subtract',
'multiply', 'divide', 'mo
'rint', 'radians', 'reciprocal', 'square', 'negative', 'fix',
'ceil', 'floor',
'trunc', 'logical_not', 'arcsinh', 'arccosh', 'arctanh',
'tensordot',
'linspace', 'expand_dims', 'tile', 'arange', 'split',
'concatenate', 'stack', 'mean',
- 'maximum', 'minimum', 'swapaxes', 'clip', 'argmax', 'std', 'var',
'indices']
+ 'maximum', 'minimum', 'swapaxes', 'clip', 'argmax', 'std', 'var',
'indices', 'copysign']
@set_module('mxnet.ndarray.numpy')
@@ -2432,3 +2432,54 @@ def indices(dimensions, dtype=_np.int32, ctx=None):
else:
raise ValueError("The dimensions must be sequence of ints")
# pylint: enable=redefined-outer-name
+
+
+@set_module('mxnet.ndarray.numpy')
+def copysign(x1, x2, out=None):
+r"""copysign(x1, x2, out=None)
+
+Change the sign of x1 to that of x2, element-wise.
+
+If `x2` is a scalar, its sign will be copied to all elements of `x1`.
+
+Parameters
+--
+x1 : ndarray or scalar
+Values to change the sign of.
+x2 : ndarray or scalar
+The sign of `x2` is copied to `x1`.
+out : ndarray or None, optional
+A location into which the result is stored. It must be of the
+right shape and right type to hold the output. If not provided
+or `None`,a freshly-allocated array is returned.
+
+Returns
+---
+out : ndarray or scalar
+The values of `x1` with the sign of `x2`.
+This is a scalar if both `x1` and `x2` are scalars.
+
+Notes
+---
+This function differs from the original `numpy.copysign
+
`_ in
+the following aspects:
+
+- ``where`` param is not supported.
+
+Examples
+
+>>> np.copysign(1.3, -1)
+-1.3
+>>> 1/np.copysign(0, 1)
+inf
+>>> 1/np.copysign(0, -1)
+-inf
+
+>>> a = np.array([-1, 0, 1])
+>>> np.copysign(a, -1.1)
+array([-1., -0., -1.])
+>>> np.copysign(a, np.arange(3)-1)
+array([-1., 0., 1.])
+"""
+return _ufunc_helper(x1, x2, _npi.copysign, _np.copysign,
_npi.copysign_scalar, _npi.rcopysign_scalar, out)
diff --git a/python/mxnet/numpy/multiarray.py b/python/mxnet/numpy/multiarray.py
index 1f8aa92..632cfad 100644
--- a/python/mxnet/numpy/multiarray.py
+++ b/python/mxnet/numpy/multiarray.py
@@ -52,7 +52,7 @@ __all__ = ['ndarray', 'empty', 'array', 'zeros', 'ones',
'full', 'add', 'subtrac
'degrees', 'log2', 'log1p', 'rint', 'radians', 'reciprocal',
'square', 'negative',
'fix', 'ceil', 'floor', 'trunc', 'logical_not', 'arcsinh',
'arccosh', 'arctanh',
'tensordot', 'linspace', 'expand_dims', 'tile', 'arange', 'split',
'concatenate',
- 'stack', 'mean', 'maximum', 'minimum', 'swapaxes', 'clip',
'argmax', 'std', 'var', 'indices']
+ 'stack', 'mean', 'maximum', 'minimum', 'swapaxes', 'clip',
'argmax', 'std', 'var', 'indices', 'copysign']
# Return code for dispatching indexing function call
_NDARRAY_UNSUPPORTED_INDEXING = -1
@@ -3935,3 +3935,54 @@ def indices(dimensions, dtype=_np.int32, ctx=None):
"""
return _mx_nd_np.indices(dimensions=dimensions, dtype=dtype, ctx=ctx)
# pylint: enable=redefined-outer-name
+
+
+@set_module('mxnet.numpy')
+def copysign(x1, x2, out=None):
+r"""copysign(x1, x2, out=None)
+
+Change the sign of x1 to that of x2, element-wise.
+
+If `x2` is a scalar, its sign will be copied to all elements of `x1`.
+
+
[GitHub] [incubator-mxnet] sxjscience merged pull request #15851: [Numpy] Numpy copysign
sxjscience merged pull request #15851: [Numpy] Numpy copysign URL: https://github.com/apache/incubator-mxnet/pull/15851 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] sxjscience commented on a change in pull request #16009: [Numpy] Numpy compatible bitwise_and operator
sxjscience commented on a change in pull request #16009: [Numpy] Numpy compatible bitwise_and operator URL: https://github.com/apache/incubator-mxnet/pull/16009#discussion_r324475447 ## File path: tests/python/unittest/test_numpy_op.py ## @@ -1080,6 +1080,53 @@ def hybrid_forward(self, F, a, *args): assert same(mx_out.asnumpy(), np_out) +@with_seed() +@use_np +def test_np_bitwise_and(): +class TestBitwiseAnd(HybridBlock): +def __init__(self): +super(TestBitwiseAnd, self).__init__() + +def hybrid_forward(self, F, x1, x2): +return F.np.bitwise_and(x1, x2) + +shapes = [ +((3, 1), (3, 1)), +((3, 1, 2), (3, 1, 2)), +((1, ),(1, )), +((3, 0), (3, 0)), # zero-size shape +((0, 1), (0, 1)), # zero-size shape +((2, 0, 2), (2, 0, 2)), # zero-size shape +((1, ), (3, )), # broadcast +((2, 3), (2, 1)), # broadcast +((1, 3), (2, 3)), # broadcast +((1, 3), (2, 0, 3)), # broadcast to zero-size shape +((1, 0, 1), (3, 0, 1)), # broadcast of zero-size shape +((), ()), # zero-dim shape +] + +for hybridize in [True, False]: +for shape in shapes: +x1_shape, x2_shape = shape + +test_bitwise_and = TestBitwiseAnd() +if hybridize: +test_bitwise_and.hybridize() + +x1 = rand_ndarray(x1_shape, dtype=_np.dtype(int)).as_np_ndarray() +x2 = rand_ndarray(x2_shape, dtype=_np.dtype(int)).as_np_ndarray() Review comment: Should we test for all the supported dtypes, I think we support uint8, int8, int32, int64. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324475346 ## File path: tests/python/unittest/test_numpy_op.py ## @@ -1853,6 +1853,111 @@ def hybrid_forward(self, F, x): assert_almost_equal(mx_ret.asnumpy(), np_ret, atol=1e-5, rtol=1e-4) +@with_seed() +@use_np +def test_np_copysign(): +class TestCopysign(HybridBlock): +def __init__(self): +super(TestCopysign, self).__init__() + +def hybrid_forward(self, F, a1, a2): + return F.np.copysign(a1, a2) + +def get_grad(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(-1, *a1.shape) +sign = _np.sum(sign, axis=0) +return sign, _np.zeros_like(a2) + +def get_grad_left(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(a1.shape) +return sign + +def get_grad_right(a1, a2): +return _np.zeros_like(a2) + +shapes = [ +(), +(1), +(2, 1), +(3, 2, 1), +(4, 3, 2, 1), +(2, 4, 3, 2, 1) +] +types = ['float16', 'float32', 'float64', 'int8', 'int32', 'int64'] +for a1shape in shapes: +for a2shape in shapes: +for hybridize in [True, False]: +for dtype in types: +test_copysign = TestCopysign() +if hybridize: +test_copysign.hybridize() +rtol = 1e-3 +atol = 1e-5 +a1_np = _np.array(_np.random.uniform(-1.0, 1.0, a1shape), dtype=dtype) +a2_np = _np.array(_np.random.uniform(-1.0, 1.0, a2shape), dtype=dtype) +a1 = np.array(a1_np, dtype=dtype) +a2 = np.array(a2_np, dtype=dtype) +a1.attach_grad() +a2.attach_grad() +expected_np = _np.copysign(a1_np, a2_np) +with mx.autograd.record(): +mx_out = test_copysign(a1, a2) +assert mx_out.shape == expected_np.shape Review comment: I agree that in the future we should do it, after more dtypes and casting is supported. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] sxjscience commented on issue #16016: [numpy] operator ravel, derive from reshape
sxjscience commented on issue #16016: [numpy] operator ravel, derive from reshape URL: https://github.com/apache/incubator-mxnet/pull/16016#issuecomment-531588040 @tingying2020 Would you make a small change in the code to trigger the CI again? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] sxjscience commented on a change in pull request #15851: [Numpy] Numpy copysign
sxjscience commented on a change in pull request #15851: [Numpy] Numpy copysign URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324474947 ## File path: tests/python/unittest/test_numpy_op.py ## @@ -1853,6 +1853,111 @@ def hybrid_forward(self, F, x): assert_almost_equal(mx_ret.asnumpy(), np_ret, atol=1e-5, rtol=1e-4) +@with_seed() +@use_np +def test_np_copysign(): +class TestCopysign(HybridBlock): +def __init__(self): +super(TestCopysign, self).__init__() + +def hybrid_forward(self, F, a1, a2): + return F.np.copysign(a1, a2) + +def get_grad(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(-1, *a1.shape) +sign = _np.sum(sign, axis=0) +return sign, _np.zeros_like(a2) + +def get_grad_left(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(a1.shape) +return sign + +def get_grad_right(a1, a2): +return _np.zeros_like(a2) + +shapes = [ +(), +(1), +(2, 1), +(3, 2, 1), +(4, 3, 2, 1), +(2, 4, 3, 2, 1) +] +types = ['float16', 'float32', 'float64', 'int8', 'int32', 'int64'] +for a1shape in shapes: +for a2shape in shapes: +for hybridize in [True, False]: +for dtype in types: +test_copysign = TestCopysign() +if hybridize: +test_copysign.hybridize() +rtol = 1e-3 +atol = 1e-5 +a1_np = _np.array(_np.random.uniform(-1.0, 1.0, a1shape), dtype=dtype) +a2_np = _np.array(_np.random.uniform(-1.0, 1.0, a2shape), dtype=dtype) +a1 = np.array(a1_np, dtype=dtype) +a2 = np.array(a2_np, dtype=dtype) +a1.attach_grad() +a2.attach_grad() +expected_np = _np.copysign(a1_np, a2_np) +with mx.autograd.record(): +mx_out = test_copysign(a1, a2) +assert mx_out.shape == expected_np.shape Review comment: Ignore my previous comments. I think we'd better check the return types after we support arbitrary dtype combinations in deepnumpy. So it's okay to not check the return types now. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] sxjscience commented on a change in pull request #15851: [Numpy] Numpy copysign
sxjscience commented on a change in pull request #15851: [Numpy] Numpy copysign URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324474508 ## File path: tests/python/unittest/test_numpy_op.py ## @@ -1853,6 +1853,111 @@ def hybrid_forward(self, F, x): assert_almost_equal(mx_ret.asnumpy(), np_ret, atol=1e-5, rtol=1e-4) +@with_seed() +@use_np +def test_np_copysign(): +class TestCopysign(HybridBlock): +def __init__(self): +super(TestCopysign, self).__init__() + +def hybrid_forward(self, F, a1, a2): + return F.np.copysign(a1, a2) + +def get_grad(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(-1, *a1.shape) +sign = _np.sum(sign, axis=0) +return sign, _np.zeros_like(a2) + +def get_grad_left(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(a1.shape) +return sign + +def get_grad_right(a1, a2): +return _np.zeros_like(a2) + +shapes = [ +(), +(1), +(2, 1), +(3, 2, 1), +(4, 3, 2, 1), +(2, 4, 3, 2, 1) +] +types = ['float16', 'float32', 'float64', 'int8', 'int32', 'int64'] +for a1shape in shapes: +for a2shape in shapes: +for hybridize in [True, False]: +for dtype in types: +test_copysign = TestCopysign() +if hybridize: +test_copysign.hybridize() +rtol = 1e-3 +atol = 1e-5 +a1_np = _np.array(_np.random.uniform(-1.0, 1.0, a1shape), dtype=dtype) +a2_np = _np.array(_np.random.uniform(-1.0, 1.0, a2shape), dtype=dtype) +a1 = np.array(a1_np, dtype=dtype) +a2 = np.array(a2_np, dtype=dtype) +a1.attach_grad() +a2.attach_grad() +expected_np = _np.copysign(a1_np, a2_np) +with mx.autograd.record(): +mx_out = test_copysign(a1, a2) +assert mx_out.shape == expected_np.shape Review comment: Test if mx_out.dtype matches expected_np.dtype. I think we can later add a utility to match the shape + dtype. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] eric-haibin-lin commented on issue #16175: [Dataset] add shard API
eric-haibin-lin commented on issue #16175: [Dataset] add shard API URL: https://github.com/apache/incubator-mxnet/pull/16175#issuecomment-531585161 @davisliang This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] eric-haibin-lin opened a new pull request #16175: [Dataset] add shard API
eric-haibin-lin opened a new pull request #16175: [Dataset] add shard API URL: https://github.com/apache/incubator-mxnet/pull/16175 ## Description ## Add an API to shard the dataset. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [ ] Feature1, tests, (and when applicable, API doc) - [ ] Feature2, tests, (and when applicable, API doc) ## Comments ## - If this change is a backward incompatible change, why must this change be made. - Interesting edge cases to note here This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add MKL-DNN Convolution
ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add
MKL-DNN Convolution
URL: https://github.com/apache/incubator-mxnet/pull/16141#discussion_r324465796
##
File path: src/operator/nn/mkldnn/mkldnn_convolution.cc
##
@@ -126,13 +127,12 @@ mkldnn::convolution_forward::primitive_desc
GetConvFwdImpl(const MKLDNNConvFullP
if (param.conv_param.dilate.ndim() == 0 && bias_md_ptr == nullptr) {
mkldnn::convolution_forward::desc desc(prop,
mkldnn::algorithm::convolution_direct, data_md,
- weight_md, out_md, strides,
padding, padding,
- mkldnn::padding_kind::zero);
Review comment:
Is `padding_kind` disabled or `zero` becomes default option from MKL-DNN
v1.0?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add MKL-DNN Convolution
ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add
MKL-DNN Convolution
URL: https://github.com/apache/incubator-mxnet/pull/16141#discussion_r324465554
##
File path: src/operator/nn/mkldnn/mkldnn_convolution-inl.h
##
@@ -79,54 +79,63 @@ struct MKLDNNConvFullParam {
MKLDNNPostEltwiseParam postsum_act_param;
};
-mkldnn::convolution_forward::primitive_desc GetConvFwdImpl(const
MKLDNNConvFullParam ,
- const bool is_train,
- const NDArray ,
- const NDArray
,
- const NDArray *bias,
- const NDArray
);
+std::shared_ptr GetConvFwdImpl(
+const ConvolutionParam , const bool is_train, const NDArray ,
const NDArray ,
+const NDArray *bias, const NDArray );
class MKLDNNConvForward {
public:
- mkldnn::convolution_forward::primitive_desc fwd_pd;
-
MKLDNNConvForward(const MKLDNNConvFullParam , const bool is_train,
const NDArray ,
-const NDArray , const NDArray *bias, const NDArray
);
+const NDArray , const NDArray *bias, const NDArray
);
- void SetNewMem(const mkldnn::memory , const mkldnn::memory ,
- const mkldnn::memory *bias, const mkldnn::memory );
+ const mkldnn::convolution_forward () const { return *fwd_; }
- void SetNewMem(const mkldnn::memory , const mkldnn::memory ) {
-this->data_->set_data_handle(data.get_data_handle());
-this->out_->set_data_handle(output.get_data_handle());
- }
-
- const mkldnn::convolution_forward () const {
-return *fwd_;
- }
+ const mkldnn::convolution_forward::primitive_desc () const { return
*pd_; }
private:
std::shared_ptr fwd_;
- std::shared_ptr data_;
- std::shared_ptr weight_;
- std::shared_ptr bias_;
- std::shared_ptr out_;
+ std::shared_ptr pd_;
};
typedef ParamOpSign MKLDNNConvSignature;
-MKLDNNConvForward (const ConvolutionParam ,
- const bool is_train, const NDArray ,
- const NDArray , const NDArray *bias,
- const NDArray );
-
void MKLDNNConvolutionForwardFullFeature(const MKLDNNConvFullParam ,
const OpContext ,
MKLDNNConvForward *fwd,
const std::vector _data,
const std::vector ,
const std::vector _data);
+void MKLDNNConvolutionForward(const nnvm::NodeAttrs ,
+ const OpContext ,
+ const std::vector _data,
+ const std::vector ,
+ const std::vector _data);
+
+class MKLDNNConvBackward {
+ public:
+ MKLDNNConvBackward(const MKLDNNConvFullParam , const NDArray ,
const NDArray ,
+ const NDArray *bias, const NDArray );
+
+ const mkldnn::convolution_backward_data () const { return
*bwd_data_; }
+
+ const mkldnn::convolution_backward_weights () const { return
*bwd_weight_; }
+
+ const mkldnn::convolution_backward_data::primitive_desc () const {
+return *bwd_data_pd_;
+ }
+
+ const mkldnn::convolution_backward_weights::primitive_desc ()
const {
+return *bwd_weights_pd_;
+ }
+
+ private:
+ std::shared_ptr
bwd_data_pd_;
+ std::shared_ptr
bwd_weights_pd_;
+ std::shared_ptr bwd_data_;
+ std::shared_ptr bwd_weight_;
+};
+
Review comment:
Please use `weight` or `weights` consistently.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add MKL-DNN Convolution
ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add
MKL-DNN Convolution
URL: https://github.com/apache/incubator-mxnet/pull/16141#discussion_r324465357
##
File path: src/operator/nn/mkldnn/mkldnn_base-inl.h
##
@@ -277,6 +277,19 @@ inline static mkldnn::memory::desc GetWeightDesc(const
NDArray ,
}
}
+inline static const std::vector GetMKLDNNInputArray(const
std::vector ) {
+ std::vector ret;
+ ret.reserve(inputs.size());
+ for (const auto : inputs) {
+if (in.IsView() && in.IsMKLDNNData()) {
+ ret.push_back(in.Reorder2Default());
+} else {
+ ret.push_back(in);
Review comment:
Return `inputs` here to avoid overhead?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add MKL-DNN Convolution
ciyongch commented on a change in pull request #16141: [mkldnn-v1.0] Add
MKL-DNN Convolution
URL: https://github.com/apache/incubator-mxnet/pull/16141#discussion_r324465659
##
File path: src/operator/nn/mkldnn/mkldnn_convolution.cc
##
@@ -57,7 +58,7 @@ mkldnn::convolution_forward::primitive_desc
GetConvFwdImpl(const MKLDNNConvFullP
auto bias_md =
bias ? (param.mkldnn_param.quantized ? GetMemDesc(*bias,
mshadow::kInt32) : GetMemDesc(*bias))
: mkldnn::memory::desc{
- {}, mkldnn::memory::data_type::data_undef,
mkldnn::memory::format::any};
+ {}, mkldnn::memory::data_type::undef,
mkldnn::memory::format_tag::any};
Review comment:
indent
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. anirudh2290 pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new ddee92e Bump the publish timestamp. ddee92e is described below commit ddee92e267a0f017a1d7088e9173640554ddbc1d Author: mxnet-ci AuthorDate: Sun Sep 15 13:35:32 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..9e01d92 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Sun Sep 15 13:35:32 UTC 2019
[GitHub] [incubator-mxnet] QueensGambit commented on issue #16173: Saving and loading cudNN autotune and graph optimization
QueensGambit commented on issue #16173: Saving and loading cudNN autotune and
graph optimization
URL:
https://github.com/apache/incubator-mxnet/issues/16173#issuecomment-531559033
Thank you for the reply @pengzhao-intel. I updated the description on MKLDNN.
I see the point about portability and backward compatibility issues.
Maybe it is better to define `optimize` as a string argument which must be
in `{'on_bind', 'save_reload', 'disabled'}`:
```python
def bind(self, ctx, args, args_grad=None, grad_req='write',
aux_states=None, group2ctx=None, shared_exec=None,
optimize='on_bind'):
"""
# ...
optimize : str, optional, default 'on_bind'
must be in {'on_bind', 'save_reload', 'disabled'}
'on_bind': Graph optimization / cuDNN autotune is
executed during model binding
'save_reload': MXNet attempts to recover previous
optimization information.
Otherwise MXNet will perform optimization
and save it to disk.
'disabled': No graph optimization / cuDNN autotune is
performed
"""
```
In the default case `optimize='on_bind'`, it will behave the same way as
currently and all previous code will behave the same.
As a different aspect, it might be preferable to treat graph optimization
(MKLDNN graph optimization / TensorRT graph fusion) as a different entity
compared to cudNN autotune because cudNN autotune might also be performed on
fused graphs in future versions.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel closed pull request #15741: MKL-DNN LBR-GRU Inference Integration (FP32 LBR-GRU)
pengzhao-intel closed pull request #15741: MKL-DNN LBR-GRU Inference Integration (FP32 LBR-GRU) URL: https://github.com/apache/incubator-mxnet/pull/15741 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15741: MKL-DNN LBR-GRU Inference Integration (FP32 LBR-GRU)
pengzhao-intel commented on issue #15741: MKL-DNN LBR-GRU Inference Integration (FP32 LBR-GRU) URL: https://github.com/apache/incubator-mxnet/pull/15741#issuecomment-531556087 closing this PR since we will migrate it with MKL-DNN 1.0. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #16141: [mkldnn-v1.0] Add MKL-DNN Convolution
pengzhao-intel commented on issue #16141: [mkldnn-v1.0] Add MKL-DNN Convolution URL: https://github.com/apache/incubator-mxnet/pull/16141#issuecomment-531555822 @TaoLv @ciyongch @ZhennanQin please help to review :) This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324453236
##
File path: python/mxnet/ndarray/numpy/_op.py
##
@@ -293,3 +293,46 @@ def power(x1, x2, out=None):
This is a scalar if both x1 and x2 are scalars.
"""
return _ufunc_helper(x1, x2, _npi.power, _np.power, _npi.power_scalar,
_npi.rpower_scalar, out)
+
+
+@set_module('mxnet.ndarray.numpy')
+def copysign(x1, x2, out=None):
Review comment:
Yes. Notes have been added in docs. See `_op.py`, `multiarray.py` and
`_symbol.py`. Thank you!
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324453173
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cc
##
@@ -182,5 +202,25 @@
MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_rpower_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute)
.set_attr("FGradient",
ElemwiseGradUseOut{"_backward_rpower_scalar"});
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_copysign_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_npi_copysign_scalar"});
+
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_rcopysign_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_npi_rcopysign_scalar"});
+
+MXNET_OPERATOR_REGISTER_BINARY(_backward_npi_copysign_scalar)
+.add_argument("scalar", "float", "scalar value")
Review comment:
Yes. Have fixed it. And Have added additional tests for scalar cases. Thank
you!
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324453143 ## File path: tests/python/unittest/test_numpy_op.py ## @@ -149,6 +149,66 @@ def hybrid_forward(self, F, a): assert same(a.grad.asnumpy(), expected_grad) +@with_seed() +@use_np +def test_np_copysign(): +class TestCopysign(HybridBlock): +def __init__(self): +super(TestCopysign, self).__init__() + +def hybrid_forward(self, F, a1, a2): + return F.np.copysign(a1, a2) + +def get_grad(a1, a2): +sign = _np.logical_or(_np.logical_and(a1 < 0, a2 < 0), + _np.logical_and(a1 >= 0, a2 >= 0)) +sign = 2 * sign.astype(int) - 1 +sign = sign.reshape(-1, *a1.shape) +sign = _np.sum(sign, axis=0) +return sign, _np.zeros_like(a2) + +shapes = [ +(), +(1), +(2, 1), +(3, 2, 1), +(4, 3, 2, 1), +(2, 4, 3, 2, 1) +] +types = ['float16', 'float32', 'float64', 'int8', 'int32', 'int64'] +for a1shape in shapes: +for a2shape in shapes: +for hybridize in [True, False]: +for dtype in types: +test_copysign = TestCopysign() +if hybridize: +test_copysign.hybridize() +rtol = 1e-3 +atol = 1e-5 +a1_np = _np.array(_np.random.uniform(-1.0, 1.0, a1shape), dtype=dtype) +a2_np = _np.array(_np.random.uniform(-1.0, 1.0, a2shape), dtype=dtype) +a1 = mx.nd.array(a1_np).as_np_ndarray() Review comment: Yes. Have fixed it. Thank you! This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324453119
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cu
##
@@ -42,6 +42,12 @@ NNVM_REGISTER_OP(_npi_mod)
NNVM_REGISTER_OP(_npi_power)
.set_attr("FCompute", BinaryBroadcastCompute);
+NNVM_REGISTER_OP(_npi_copysign)
+.set_attr("FCompute", BinaryBroadcastCompute);
+
+NNVM_REGISTER_OP(_backward_npi_copysign)
+.set_attr("FCompute", BinaryBroadcastBackwardUseIn);
Review comment:
fixed.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
hzfan commented on a change in pull request #15851: [Numpy] Numpy copysign
URL: https://github.com/apache/incubator-mxnet/pull/15851#discussion_r324453107
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cc
##
@@ -182,5 +202,25 @@
MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_rpower_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute)
.set_attr("FGradient",
ElemwiseGradUseOut{"_backward_rpower_scalar"});
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_copysign_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_npi_copysign_scalar"});
+
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_rcopysign_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_npi_rcopysign_scalar"});
+
+MXNET_OPERATOR_REGISTER_BINARY(_backward_npi_copysign_scalar)
+.add_argument("scalar", "float", "scalar value")
+.set_attr_parser([](NodeAttrs *attrs) { attrs->parsed =
std::stod(attrs->dict["scalar"]); })
+.set_attr("FCompute",
+BinaryScalarOp::Backward);
+
+MXNET_OPERATOR_REGISTER_BINARY(_backward_npi_rcopysign_scalar)
+.add_argument("scalar", "float", "scalar value")
+.set_attr_parser([](NodeAttrs *attrs) { attrs->parsed =
std::stod(attrs->dict["scalar"]); })
+.set_attr("FCompute", BinaryScalarOp::Backward<
+ cpu, mshadow_op::rcopysign_grad>);
Review comment:
fixed.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[incubator-mxnet-site] branch asf-site updated: Bump the publish timestamp.
This is an automated email from the ASF dual-hosted git repository. anirudh2290 pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git The following commit(s) were added to refs/heads/asf-site by this push: new 8ab3b60 Bump the publish timestamp. 8ab3b60 is described below commit 8ab3b605387b406ebacb81752a68f7f66d7e0350 Author: mxnet-ci AuthorDate: Sun Sep 15 07:36:49 2019 + Bump the publish timestamp. --- date.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/date.txt b/date.txt new file mode 100644 index 000..a225561 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Sun Sep 15 07:36:49 UTC 2019
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #16167: [RFC] Apache MXNet 2.0 Roadmap
pengzhao-intel commented on issue #16167: [RFC] Apache MXNet 2.0 Roadmap URL: https://github.com/apache/incubator-mxnet/issues/16167#issuecomment-531542441 @szha Really great proposal and we may want to add some items in 2.0 too. Is there a timeline of 2.0? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #16173: Saving and loading cudNN autotune and graph optimization
pengzhao-intel commented on issue #16173: Saving and loading cudNN autotune and graph optimization URL: https://github.com/apache/incubator-mxnet/issues/16173#issuecomment-531542295 FYI, MKLDNN graph fusion is already enabled by default :) One more, saving the fused graph maybe cause portable issues and break backward compatible so we need a solution to fall back as well. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] shdyn opened a new issue #16174: How to build new operators?
shdyn opened a new issue #16174: How to build new operators? URL: https://github.com/apache/incubator-mxnet/issues/16174 Hi, I need to compile the operators in https://github.com/deepinsight/insightface/tree/master/3rdparty/operator. I tried to copy files in that folder to incubator-mxnet/src/operator and compile the source codes using incubator-mxnet/docs/install/install_mxnet_ubuntu_python.sh. Then I install these operators using incubator-mxnet/python/setup.py. However, I still cannot run these operators? Any suggestions. Thanks in advance! This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on a change in pull request #16124: [numpy] [tvm] operator true_divide
reminisce commented on a change in pull request #16124: [numpy] [tvm] operator
true_divide
URL: https://github.com/apache/incubator-mxnet/pull/16124#discussion_r324447425
##
File path: src/operator/numpy/np_true_divide.cc
##
@@ -41,19 +46,97 @@ bool TrueDivideType(const nnvm::NodeAttrs& attrs,
const int lhs_dtype = in_attrs->at(0);
const int rhs_dtype = in_attrs->at(1);
CHECK_EQ(lhs_dtype, rhs_dtype)
-<< "_true_divide currently only supports same dtype for dividend and
divisor";
+ << "_true_divide currently only supports same dtype for dividend and
divisor";
}
- auto is_float = [](const int dtype) {
-return dtype == mshadow::kFloat32 || dtype == mshadow::kFloat64 || dtype
== mshadow::kFloat16;
- };
-
- for (const int dtype : *in_attrs) {
-CHECK(is_float(dtype)) << "_true_divide currently only supports float
dtype";
+ if (IsIntType(in_attrs->at(0))) {
+TYPE_ASSIGN_CHECK(*out_attrs, 0, mshadow::kFloat64);
Review comment:
We use `float32` as default dtype.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on a change in pull request #16124: [numpy] [tvm] operator true_divide
reminisce commented on a change in pull request #16124: [numpy] [tvm] operator
true_divide
URL: https://github.com/apache/incubator-mxnet/pull/16124#discussion_r324447266
##
File path: contrib/tvmop/core/umath.py
##
@@ -0,0 +1,112 @@
+ # Licensed to the Apache Software Foundation (ASF) under one
+ # or more contributor license agreements. See the NOTICE file
+ # distributed with this work for additional information
+ # regarding copyright ownership. The ASF licenses this file
+ # to you under the Apache License, Version 2.0 (the
+ # "License"); you may not use this file except in compliance
+ # with the License. You may obtain a copy of the License at
+ #
+ # http://www.apache.org/licenses/LICENSE-2.0
+ #
+ # Unless required by applicable law or agreed to in writing,
+ # software distributed under the License is distributed on an
+ # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ # KIND, either express or implied. See the License for the
+ # specific language governing permissions and limitations
+ # under the License.
+import tvm
+from .. import defop, AllTypes, AllTypesButHalf
+
+def compute_true_divide(dtype, ndim):
+A = tvm.placeholder([tvm.var() for _ in range(ndim)], name='A',
dtype=dtype)
+B = tvm.placeholder([tvm.var() for _ in range(ndim)], name='B',
dtype=dtype)
+if dtype in ["float16", "float32", "float64"]:
+C = tvm.compute([tvm.var() for _ in range(ndim)],
+lambda *index: A[index] / B[index], name='C')
+else:
+C = tvm.compute([tvm.var() for _ in range(ndim)],
+lambda *index: A[index].astype("float64") /
Review comment:
This always produces fp64 dtype as output. We should support allowing users
to set desired float dtypes. Also note that in numpy (divide/true_divide), it's
not allowed to set integer dtype for outputs.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on a change in pull request #16124: [numpy] [tvm] operator true_divide
reminisce commented on a change in pull request #16124: [numpy] [tvm] operator
true_divide
URL: https://github.com/apache/incubator-mxnet/pull/16124#discussion_r324447266
##
File path: contrib/tvmop/core/umath.py
##
@@ -0,0 +1,112 @@
+ # Licensed to the Apache Software Foundation (ASF) under one
+ # or more contributor license agreements. See the NOTICE file
+ # distributed with this work for additional information
+ # regarding copyright ownership. The ASF licenses this file
+ # to you under the Apache License, Version 2.0 (the
+ # "License"); you may not use this file except in compliance
+ # with the License. You may obtain a copy of the License at
+ #
+ # http://www.apache.org/licenses/LICENSE-2.0
+ #
+ # Unless required by applicable law or agreed to in writing,
+ # software distributed under the License is distributed on an
+ # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ # KIND, either express or implied. See the License for the
+ # specific language governing permissions and limitations
+ # under the License.
+import tvm
+from .. import defop, AllTypes, AllTypesButHalf
+
+def compute_true_divide(dtype, ndim):
+A = tvm.placeholder([tvm.var() for _ in range(ndim)], name='A',
dtype=dtype)
+B = tvm.placeholder([tvm.var() for _ in range(ndim)], name='B',
dtype=dtype)
+if dtype in ["float16", "float32", "float64"]:
+C = tvm.compute([tvm.var() for _ in range(ndim)],
+lambda *index: A[index] / B[index], name='C')
+else:
+C = tvm.compute([tvm.var() for _ in range(ndim)],
+lambda *index: A[index].astype("float64") /
Review comment:
This always produces fp64 dtype as output. We should support allowing users
to set desired output dtype.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] reminisce closed pull request #16116: np.ndarray.repeat support & doc for np.repeat
reminisce closed pull request #16116: np.ndarray.repeat support & doc for np.repeat URL: https://github.com/apache/incubator-mxnet/pull/16116 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on issue #16116: np.ndarray.repeat support & doc for np.repeat
reminisce commented on issue #16116: np.ndarray.repeat support & doc for np.repeat URL: https://github.com/apache/incubator-mxnet/pull/16116#issuecomment-531537737 Fixed in https://github.com/apache/incubator-mxnet/pull/16157 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
