[GitHub] jasonyu1996 opened a new issue #12390: [Feature Request] dtype for modules
jasonyu1996 opened a new issue #12390: [Feature Request] dtype for modules URL: https://github.com/apache/incubator-mxnet/issues/12390 It seems that only a very small number of modules among all the modules available in the library supports specifying a `dtype` in their constructors (including `Dense` and `Embedding`). If not specified, the default parameters would be `float32`, which could be a troublesome limitation in some cases. On the forum: https://discuss.mxnet.io/t/default-datatype/1762 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416130369 the machines are already sshable without password between each other This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416125937 Thank U, But if installed mxnet by pip, How to use python launch.py since I didn't download the mxnet source code? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416130112 I download the mxnet source code, but didn't compile since i used the mxnet installed by pip. when i use the launch.py provided in the mxnet source code, some errors shows up: xiaomin.wu@iva0605:/autofs/data56/public/xiaomin.wu/code/dist_mxnet_test$ python /autofs/data56/public/xiaomin.wu/software/incubator-mxnet-1.2.0/tools/launch.py -n 2 -s 2 -H hosts --sync-dst-dir /home/xiaomin.wu/cifar10_dist --launcher ssh "python /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py" Can't load dmlc_tracker package. Perhaps you need to run git submodule update --init --recursive Traceback (most recent call last): File "/autofs/data56/public/xiaomin.wu/software/incubator-mxnet-1.2.0/tools/launch.py", line 128, in main() File "/autofs/data56/public/xiaomin.wu/software/incubator-mxnet-1.2.0/tools/launch.py", line 96, in main args = dmlc_opts(args) File "/autofs/data56/public/xiaomin.wu/software/incubator-mxnet-1.2.0/tools/launch.py", line 48, in dmlc_opts from dmlc_tracker import opts ImportError: No module named dmlc_tracker This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416132996 I don't think so because I tried: python /home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/mxnet/tools/launch.py -n 2 -s 2 -H hosts --sync-dst-dir /home/xiaomin.wu/cifar10_dist --launcher ssh "python /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py" the same errors show up This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416146366 I update the version to 1.2.1 by pip, but new error shows up:( xiaomin.wu@iva0605:/autofs/data56/public/xiaomin.wu/code/dist_mxnet_test$ python /home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/mxnet/tools/launch.py -n 2 -s 2 -H hosts --sync-dst-dir /home/xiaomin.wu/cifar10_dist --launcher ssh "python /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py" 2018-08-27 15:54:12,883 INFO rsync /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/ -> 10.14.6.5:/home/xiaomin.wu/cifar10_dist xiaomin.wu@10.14.6.5's password: /home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`. from ._conv import register_converters as _register_converters 2018-08-27 15:54:18,462 INFO rsync /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/ -> 10.14.6.8:/home/xiaomin.wu/cifar10_dist xiaomin.wu@10.14.6.5's password: xiaomin.wu@10.14.6.5's password: Traceback (most recent call last): File "/autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py", line 23, in import mxnet as mx ImportError: No module named mxnet Traceback (most recent call last): File "/autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py", line 23, in import mxnet as mx ImportError: No module named mxnet Exception in thread Thread-5: Traceback (most recent call last): File "/home/xiaomin.wu/anaconda2/lib/python2.7/threading.py", line 801, in __bootstrap_inner self.run() File "/home/xiaomin.wu/anaconda2/lib/python2.7/threading.py", line 754, in run self.__target(*self.__args, **self.__kwargs) File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/dmlc_tracker/ssh.py", line 61, in run subprocess.check_call(prog, shell = True) File "/home/xiaomin.wu/anaconda2/lib/python2.7/subprocess.py", line 186, in check_call raise CalledProcessError(retcode, cmd) CalledProcessError: Command 'ssh -o StrictHostKeyChecking=no 10.14.6.8 -p 22 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64::/usr/local/cuda-8.0/lib64:~/TensorRT-4.0.0.3/lib; export DMLC_ROLE=worker; export DMLC_PS_ROOT_PORT=9093; export DMLC_PS_ROOT_URI=10.14.6.5; export DMLC_NUM_SERVER=2; export DMLC_NUM_WORKER=2; cd /home/xiaomin.wu/cifar10_dist; python /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py'' returned non-zero exit status 1 Exception in thread Thread-3: Traceback (most recent call last): File "/home/xiaomin.wu/anaconda2/lib/python2.7/threading.py", line 801, in __bootstrap_inner self.run() File "/home/xiaomin.wu/anaconda2/lib/python2.7/threading.py", line 754, in run self.__target(*self.__args, **self.__kwargs) File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/dmlc_tracker/ssh.py", line 61, in run subprocess.check_call(prog, shell = True) File "/home/xiaomin.wu/anaconda2/lib/python2.7/subprocess.py", line 186, in check_call raise CalledProcessError(retcode, cmd) CalledProcessError: Command 'ssh -o StrictHostKeyChecking=no 10.14.6.8 -p 22 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64::/usr/local/cuda-8.0/lib64:~/TensorRT-4.0.0.3/lib; export DMLC_ROLE=server; export DMLC_PS_ROOT_PORT=9093; export DMLC_PS_ROOT_URI=10.14.6.5; export DMLC_NUM_SERVER=2; export DMLC_NUM_WORKER=2; cd /home/xiaomin.wu/cifar10_dist; python /autofs/data56/public/xiaomin.wu/code/dist_mxnet_test/cifar10_dist.py'' returned non-zero exit status 1 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416823724 @eric-haibin-lin 我可以从a无密码ssh到b，也可以从b无密码ssh到a，但不能a无密码ssh到a，b也不能无密码ssh到b。而问题应该正是如此 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416133087 the vesion of the mxnet i installed buy pip is 1.2.0 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee removed a comment on issue #12363: distributed training notebook tests
meanmee removed a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416158008 Permission denied, please try again. xiaomin.wu@10.14.6.5's password: Permission denied, please try again. xiaomin.wu@10.14.6.5's password: Permission denied (publickey,password). rsync: connection unexpectedly closed (0 bytes received so far) [sender] rsync error: unexplained error (code 255) at io.c(226) [sender=3.1.1] Traceback (most recent call last): File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/mxnet/tools/launch.py", line 128, in main() File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/mxnet/tools/launch.py", line 113, in main ssh.submit(args) File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/dmlc_tracker/ssh.py", line 86, in submit hostIP=args.host_ip) File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/dmlc_tracker/tracker.py", line 428, in submit fun_submit(nworker, nserver, envs) File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/dmlc_tracker/ssh.py", line 69, in ssh_submit sync_dir(local_dir, h, working_dir) File "/home/xiaomin.wu/anaconda2/lib/python2.7/site-packages/dmlc_tracker/ssh.py", line 21, in sync_dir subprocess.check_call([prog], shell = True) File "/home/xiaomin.wu/anaconda2/lib/python2.7/subprocess.py", line 186, in check_call raise CalledProcessError(retcode, cmd) subprocess.CalledProcessError: Command '['rsync -az --rsh="ssh -o StrictHostKeyChecking=no -p 22" /home/xiaomin.wu/cifar10_dist/ 10.14.6.5:/home/xiaomin.wu/cifar10_dist']' returned non-zero exit status 255 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] jinhuang415 closed issue #10175: MXNet MKLDNN build dependency/flow discussion
jinhuang415 closed issue #10175: MXNet MKLDNN build dependency/flow discussion URL: https://github.com/apache/incubator-mxnet/issues/10175 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] meanmee commented on issue #12363: distributed training notebook tests
meanmee commented on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416823724 @eric-haibin-lin 我可以从a无密码ssh到b，也可以从b无密码ssh到a，但不能a无密码ssh到a，b也不能无密码ssh到b。而问题应该正是如此 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] yuxiangw opened a new issue #12389: [Bug] Not able to detect out of bound index on an ndarray. potential memory overflow.
yuxiangw opened a new issue #12389: [Bug] Not able to detect out of bound index
on an ndarray. potential memory overflow.
URL: https://github.com/apache/incubator-mxnet/issues/12389
## Minimum reproducible example
'''
from mxnet import ndarray as nd
import numpy as np
# Declare an mxnet ndarray
x = nd.array(range(5))
print(x)
# Try an out of bound index
idx = [5]
print('mxnet:',x[idx])
'''
## Output:
'''
[0. 1. 2. 3. 4.]
mxnet:
[-3.689349e+19]
'''
Note that there is no complaint what-so-ever. Not even a warning.
On the other hand, if you do the following instead:
'''
# indexing with integer
idx=5
x[idx]
'''
Then you get the desired error message:
'''
IndexError: index 5 is out of bounds for axis 0 with size 5
'''
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] zheng-da closed pull request #12151: fix a minor bug in while_loop
zheng-da closed pull request #12151: fix a minor bug in while_loop
URL: https://github.com/apache/incubator-mxnet/pull/12151
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/python/mxnet/symbol/contrib.py b/python/mxnet/symbol/contrib.py
index 38195bd62ff..f89c73164fe 100644
--- a/python/mxnet/symbol/contrib.py
+++ b/python/mxnet/symbol/contrib.py
@@ -539,9 +539,6 @@ def _union_inputs(*graphs):
# find symbols used in either cond_g or func_g
input_syms, ((cond_input_locs, _), (func_input_locs, func_var_locs)) = \
_union_inputs(cond_g, func_g)
-for i_th, loc in enumerate(func_var_locs, 1):
-if loc == -1:
-raise ValueError("The %d-th loop_var doesn't involve into the
computation" % i_th)
result = symbol._internal._while_loop(
# [cond, func_g, *input_syms]
cond_g,
diff --git a/tests/python/unittest/test_contrib_control_flow.py
b/tests/python/unittest/test_contrib_control_flow.py
index a4b794c9595..7205b55ec52 100644
--- a/tests/python/unittest/test_contrib_control_flow.py
+++ b/tests/python/unittest/test_contrib_control_flow.py
@@ -139,6 +139,30 @@ def hybrid_forward(self, F, *loop_vars):
assert result_s.asscalar() == 0
+def test_while_loop2():
+class TestBlock(gluon.HybridBlock):
+def __init__(self, prefix=None, params=None):
+super(TestBlock, self).__init__(prefix=prefix, params=params)
+
+# In this test, body_func only accesses one of the states,
+# so not all loop variables are used.
+def hybrid_forward(self, F, data):
+def cond_func(state1, state2):
+return state1 > 0
+def body_func(state1, state2):
+return (state2, [state2 + 1, state2 + 2])
+return F.contrib.while_loop(
+cond=cond_func,
+func=body_func,
+loop_vars=[data, data + 1],
+max_iterations=10)
+
+block = TestBlock()
+block.initialize(ctx=default_context())
+block.hybridize()
+block(mx.nd.ones((1)))
+
+
def _verify_while_loop(cond, func, loop_var_shapes, free_var_shapes, is_train,
max_iterations, is_for, n_steps):
def _create_vars(num, prefix):
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] zheng-da closed issue #11448: fail to fall back when sparse arrays are passed to MKLDNN-enabled operators.
zheng-da closed issue #11448: fail to fall back when sparse arrays are passed to MKLDNN-enabled operators. URL: https://github.com/apache/incubator-mxnet/issues/11448 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] aaronmarkham commented on issue #12102: site-wide social include
aaronmarkham commented on issue #12102: site-wide social include URL: https://github.com/apache/incubator-mxnet/pull/12102#issuecomment-416817078 @nswamy - Do you still request changes? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] pengzhao-intel commented on issue #11448: fail to fall back when sparse arrays are passed to MKLDNN-enabled operators.
pengzhao-intel commented on issue #11448: fail to fall back when sparse arrays are passed to MKLDNN-enabled operators. URL: https://github.com/apache/incubator-mxnet/issues/11448#issuecomment-416816945 @zheng-da related PR are merged. Could you verify if all issues are fixed and close the issue? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] lanking520 commented on a change in pull request #12387: MXNET-873 - Bring Clojure Package Inline with New DataDesc and Layout in Scala Package
lanking520 commented on a change in pull request #12387: MXNET-873 - Bring Clojure Package Inline with New DataDesc and Layout in Scala Package URL: https://github.com/apache/incubator-mxnet/pull/12387#discussion_r213536814 ## File path: contrib/clojure-package/examples/pre-trained-models/src/pre_trained_models/predict_image.clj ## @@ -92,7 +92,7 @@ (comment - (predict "https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/python/predict_image/cat.jpg;) + (predict "https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/python/predict_image/cat.jpg; true) Review comment: What does this `true` mean in here? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] lanking520 commented on a change in pull request #12387: MXNET-873 - Bring Clojure Package Inline with New DataDesc and Layout in Scala Package
lanking520 commented on a change in pull request #12387: MXNET-873 - Bring
Clojure Package Inline with New DataDesc and Layout in Scala Package
URL: https://github.com/apache/incubator-mxnet/pull/12387#discussion_r213537200
##
File path: contrib/clojure-package/src/org/apache/clojure_mxnet/symbol.clj
##
@@ -144,7 +144,7 @@
which must be known from the rest of the net."
([start {:keys [step repeat dtype]
:or {step (float 1) repeat (int 1) dtype base/MX_REAL_TYPE}
- :as opts}]
+ :as opts}]
Review comment:
space issue...
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] liangxi627 closed issue #12120: compile error in "image_aug_default.cc"
liangxi627 closed issue #12120: compile error in "image_aug_default.cc" URL: https://github.com/apache/incubator-mxnet/issues/12120 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] liangxi627 commented on issue #11890: Error During Make - undefined reference
liangxi627 commented on issue #11890: Error During Make - undefined reference URL: https://github.com/apache/incubator-mxnet/issues/11890#issuecomment-416811231 @sandeep-krishnamurthy Please refer to the following links. http://answers.opencv.org/question/95649/error-in-imwrite-code-specified-in-the-opencv-api-reference/ https://stackoverflow.com/questions/24439548/opencv-tutorial-load-and-display-an-image-codeblocks-fedora20 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] eric-haibin-lin commented on issue #12363: distributed training notebook tests
eric-haibin-lin commented on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416810947 @meanmee looks like there's some problem connecting to your remote instance. Did you setup passwordless ssh? I suggest you move the question to discuss.mxnet.io which is monitored actively. Github issue is more for issue/bug report or task/feature requests. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] eric-haibin-lin commented on issue #12157: Subgraph API for integrating accelerators with MXNet
eric-haibin-lin commented on issue #12157: Subgraph API for integrating accelerators with MXNet URL: https://github.com/apache/incubator-mxnet/pull/12157#issuecomment-416810308 LGTM This is an automated message from the Apache Git Service. To respond to the message, please log on 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] ghgggg closed issue #12267: some errors when extracting feature with mxnet c++ api in windows10 in gpu mode
gh closed issue #12267: some errors when extracting feature with mxnet c++ api in windows10 in gpu mode URL: https://github.com/apache/incubator-mxnet/issues/12267 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] StephanieYuan commented on issue #12376: [MXNET-854] SVRG Optimization in Python Module API
StephanieYuan commented on issue #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#issuecomment-416799044 Move svrg_optimization to python/contrib package. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] StephanieYuan edited a comment on issue #12376: [MXNET-854] SVRG Optimization in Python Module API
StephanieYuan edited a comment on issue #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#issuecomment-416799044 Moved svrg_optimization to python/contrib package. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] zheng-da edited a comment on issue #12269: get memory error when running a model exported from gluon model zoo
zheng-da edited a comment on issue #12269: get memory error when running a model exported from gluon model zoo URL: https://github.com/apache/incubator-mxnet/issues/12269#issuecomment-416798792 I can create an issue, so hopefully someone can address it This is an automated message from the Apache Git Service. To respond to the message, please log on 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] zheng-da commented on issue #12269: get memory error when running a model exported from gluon model zoo
zheng-da commented on issue #12269: get memory error when running a model exported from gluon model zoo URL: https://github.com/apache/incubator-mxnet/issues/12269#issuecomment-416798792 I don't. I can create an issue, so hopefully someone can address it This is an automated message from the Apache Git Service. To respond to the message, please log on 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] jasonyu1996 edited a comment on issue #12327: [Feature Request] support of diag for N-d arrays
jasonyu1996 edited a comment on issue #12327: [Feature Request] support of diag for N-d arrays URL: https://github.com/apache/incubator-mxnet/issues/12327#issuecomment-416790602 Hi! Thank you for your response! I just paid a visit to the numpy interfaces for computing the diagonal, and noticed that besides `numpy.diag` (which is exactly where our the design of our `diag` operator comes from) numpy provides a second diagonal extracting function `numpy.diagonal` (https://www.numpy.org/devdocs/reference/generated/numpy.diagonal.html), which in my opinion is a good reference for extending the functionality of our `diag` operator (I also noticed that `numpy.trace` also supports N-d arrays and accepts the same set of arguments as `numpy.diagonal`). However, I am not sure whether a new operator should be added or not. I wonder why numpy provides two functions, one strictly weaker than the other, which do the same thing. As for the implementation detail, I have to admit that I am not familiar with this and am therefore not sure about the possibility of further improving the performance by implementing it in ways other than simply fusing some high-level function calls together. I think it is possibly necessary to refer to the implementation of the 2-d case, which it seems does not depend on other high-level function calls (`diag` for 2-d arrays can also be implemented with an `arange` followed by a `pick`). This is an automated message from the Apache Git Service. To respond to the message, please log on 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] jasonyu1996 commented on issue #12327: [Feature Request] support of diag for N-d arrays
jasonyu1996 commented on issue #12327: [Feature Request] support of diag for N-d arrays URL: https://github.com/apache/incubator-mxnet/issues/12327#issuecomment-416790602 Hi! Thank you for your response! I just paid a visit to the numpy interfaces for computing the diagonal, and noticed that besides `numpy.diag` (which is exactly where our the design of our `diag` operator comes from) numpy provides a second diagonal extracting function `numpy.diagonal` (https://www.numpy.org/devdocs/reference/generated/numpy.diagonal.html), which in my opinion is a good reference for extending the functionality of our `diag` operator. However, I am not sure whether a new operator should be added or not. I wonder why numpy provides two functions, one strictly weaker than the other, which do the same thing. As for the implementation detail, I have to admit that I am not familiar with this and am therefore not sure about the possibility of further improving the performance by implementing it in ways other than simply fusing some high-level function calls together. I think it is possibly necessary to refer to the implementation of the 2-d case, which it seems does not depend on other high-level function calls (`diag` for 2-d arrays can also be implemented with an `arange` followed by a `pick`). This is an automated message from the Apache Git Service. To respond to the message, please log on 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] pengzhao-intel commented on issue #12269: get memory error when running a model exported from gluon model zoo
pengzhao-intel commented on issue #12269: get memory error when running a model exported from gluon model zoo URL: https://github.com/apache/incubator-mxnet/issues/12269#issuecomment-416790167 @zheng-da do you have a plan to improve it? This is an automated message from the Apache Git Service. To respond to the message, please log on 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. zhasheng 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 5ac893f Bump the publish timestamp. 5ac893f is described below commit 5ac893f88c7723e087f0494dd11feb74d52b Author: mxnet-ci AuthorDate: Wed Aug 29 00:55:53 2018 + 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..ca60c5c --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Wed Aug 29 00:55:53 UTC 2018
[GitHub] aaronmarkham opened a new pull request #12388: Installation instructions consolidation
aaronmarkham opened a new pull request #12388: Installation instructions consolidation URL: https://github.com/apache/incubator-mxnet/pull/12388 ## Description ## Different installation pages were out of date or out of sync. The install page was over-complicated. The build from source instructions weren't really working. This PR consolidates install info, directs users to where the more recent updates happen to be, and simplifies the basic user install. I started off just trying to fix the C++ setup info, but found so many other issues, I worked on those too. ## Features * Fixes a bug that made it so you couldn't preview the dev version's install page without it redirecting you. (options.js) * Updates the [build_from_source](http://34.201.8.176/versions/cplusplus_install/install/build_from_source.html) page - Point to the OS-focused pages where this information is most up-to-date. - Added a [table that maps links to each binding](http://34.201.8.176/versions/cplusplus_install/install/build_from_source.html#installing-mxnet-language-bindings). - Leaves the centos/other linux sections there since this info isn't covered elsewhere - Leaves the NCCL section alone (is this even current info?) - Links this page in the [Download Source section](http://34.201.8.176/versions/cplusplus_install/install/index.html) found at the bottom of install * Creates a new [C++ setup page](http://34.201.8.176/versions/cplusplus_install/install/c_plus_plus.html) that brings together C++ info found around the project. * Refactors the install page. - Moves the [validation section to a new page](http://34.201.8.176/versions/cplusplus_install/install/validate_mxnet.html) - Removes virtualenv (this was repeated multiple times in the file w/o much real value - Super simplifies the pip install sections - Links back to each OS's detailed setup guides - Links to specific install guide sections for Perl and Julia - and more... the file went from 2531 lines to 1062. So much easier to deal with... probably could take more pruning... * Fixes some issues on the OS guide pages - the Python binding steps were missing ## Preview http://34.201.8.176/versions/cplusplus_install/install/index.html This is an automated message from the Apache Git Service. To respond to the message, please log on 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] samskalicky edited a comment on issue #12327: [Feature Request] support of diag for N-d arrays
samskalicky edited a comment on issue #12327: [Feature Request] support of diag for N-d arrays URL: https://github.com/apache/incubator-mxnet/issues/12327#issuecomment-416780686 Hi @jasonyu1996. Thanks for reporting this. Ive been looking to implement a trace operator in mxnet per this request #10500. In preparing for implementing trace (which is really just summing the diagonal of the matix) I also noticed the limited implementation of the diag operator. Given that many MXNet users have data in the form of WxHxC (width, height, channel), and then add a 4th dimension for number/batch, what are your thoughts on general N-dimensionality approach for this operator? Is it necessary to support the general case? And how about implementation of the diag operator, as you mentioned there are already existing and high performance implementation for each sub-computation required. Do you think there is opportunity for further performance improvement (memory, time, etc.) by fusing these together? Or do you think it would be best to just implement diag calling these sub-computations separately (inside the diag operator)? Let me know if you have thoughts on this. I would be interested in working with you to implement this as well. Heres the original diag issue: #9253 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] samskalicky commented on issue #12327: [Feature Request] support of diag for N-d arrays
samskalicky commented on issue #12327: [Feature Request] support of diag for N-d arrays URL: https://github.com/apache/incubator-mxnet/issues/12327#issuecomment-416780686 Hi @jasonyu1996. Thanks for reporting this. Ive been looking to implement a trace operator in mxnet per this request #10500. In preparing for implementing trace (which is really just summing the diagonal of the matix) I also noticed the limited implementation of the diag operator. Given that many MXNet users have data in the form of WxHxC (width, height, channel), and then add a 4th dimension for number/batch, what are your thoughts on general N-dimensionality approach for this operator? Is it necessary to support the general case? And how about implementation of the diag operator, as you mentioned there are already existing and high performance implementation for each sub-computation required. Do you think there is opportunity for further performance improvement (memory, time, etc.) by fusing these together? Or do you think it would be best to just implement diag calling these sub-computations separately (inside the diag operator)? Let me know if you have thoughts on this. I would be interested in working with you to implement this as well. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] ankkhedia edited a comment on issue #12162: Edit shape.array doc and some style improvements
ankkhedia edited a comment on issue #12162: Edit shape.array doc and some style improvements URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416779364 looks good to me! Should be good to go after you retriever CI. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] ankkhedia edited a comment on issue #12162: Edit shape.array doc and some style improvements
ankkhedia edited a comment on issue #12162: Edit shape.array doc and some style improvements URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416779364 looks good to me! Should be good to go after you retrigger CI. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] ankkhedia commented on issue #12162: Edit shape.array doc and some style improvements
ankkhedia commented on issue #12162: Edit shape.array doc and some style improvements URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416779364 looks good to me! This is an automated message from the Apache Git Service. To respond to the message, please log on 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] haojin2 commented on issue #8866: src/operator/./bilinear_sampler-inl.h:105: Have not implemented the data req combinations! gdata_req=0 ggrid_req=1
haojin2 commented on issue #8866: src/operator/./bilinear_sampler-inl.h:105: Have not implemented the data req combinations! gdata_req=0 ggrid_req=1 URL: https://github.com/apache/incubator-mxnet/issues/8866#issuecomment-416778608 A fix delivered in #12386, but corresponding unit test currently blocked by the flakiness of test_bilinear_sampler This is an automated message from the Apache Git Service. To respond to the message, please log on 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] hetong007 commented on issue #12162: Edit shape.array doc and some style improvements
hetong007 commented on issue #12162: Edit shape.array doc and some style improvements URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416778171 @terrytangyuan You can trigger the CI again, with an empty commit by `git commit --allow-empty -m "Trigger CI"` This is an automated message from the Apache Git Service. To respond to the message, please log on 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] ankkhedia commented on issue #12310: Flaky test: test_ndarray.test_order
ankkhedia commented on issue #12310: Flaky test: test_ndarray.test_order URL: https://github.com/apache/incubator-mxnet/issues/12310#issuecomment-416777661 @sxjscience The issue seems to be only with ret_type = "mask" for topk operator This is an automated message from the Apache Git Service. To respond to the message, please log on 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] sxjscience commented on issue #12310: Flaky test: test_ndarray.test_order
sxjscience commented on issue #12310: Flaky test: test_ndarray.test_order URL: https://github.com/apache/incubator-mxnet/issues/12310#issuecomment-416775348 Thanks for reporting this. I find we can use other dtypes ```python import mxnet as mx import numpy as np dat_size=5 dtype=np.int32 a_npy= np.arange(dat_size ** 4, dtype=dtype).reshape((dat_size, dat_size, dat_size, dat_size)) a_nd = mx.nd.array(a_npy, ctx=mx.gpu(0), dtype=dtype) nd_ret_topk = mx.nd.topk(a_nd, axis=1, k=2, ret_typ="mask", is_ascend=False) print(nd_ret_topk.dtype) print(nd_ret_topk) ``` I'm looking for the bug in the code. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] gigasquid commented on issue #12387: MXNET-873 - Bring Clojure Package inline with new DataDesc and Layout in Scala Package
gigasquid commented on issue #12387: MXNET-873 - Bring Clojure Package inline with new DataDesc and Layout in Scala Package URL: https://github.com/apache/incubator-mxnet/pull/12387#issuecomment-416772699 @lanking520 please give a look when you get a chance This is an automated message from the Apache Git Service. To respond to the message, please log on 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] gigasquid opened a new pull request #12387: MXNET-873 - Bring Clojure Package inline with new DataDesc and Layout in Scala Package
gigasquid opened a new pull request #12387: MXNET-873 - Bring Clojure Package inline with new DataDesc and Layout in Scala Package URL: https://github.com/apache/incubator-mxnet/pull/12387 ## Description ## The Scala package has updated the DataDesc to include Layout. The Clojure package has been updated to move inline with it. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [X] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [X] Changes are complete (i.e. I finished coding on this PR) - [X] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [X] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [X] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - A layout namespace has been created and used in DataDesc - The Clojure package no longer infers the layout from the shape. It handles it the same way as the Scala package by having it be undefined - The original `module.fit` interop code has been restored. The work arounds required by the old DataDesc has been resolved with this PR https://github.com/apache/incubator-mxnet/pull/11844 - Most of the examples have moved to use `provide-data-desc` instead of `provide-data` - Formatting fixes - Tweak to Char RNN example to run for less epochs before showing the pre-loaded result This is an automated message from the Apache Git Service. To respond to the message, please log on 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] reminisce commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
reminisce commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213503117
##
File path: src/operator/subgraph/partition_graph.cc
##
@@ -0,0 +1,774 @@
+/*
+ * 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.
+ */
+
+/*!
+ * Copyright (c) 2018 by Contributors
+ * \file partition_graph.cc
+ * \brief
+ */
+#include
+#include
+#include
+#include
+#include
+#include
+
+#include "./subgraph_property.h"
+
+namespace nnvm {
+NodePtr CreateVariableNode(const std::string& name);
+}
+
+namespace mxnet {
+
+namespace op {
+
+using nnvm::Symbol;
+using nnvm::Node;
+using nnvm::NodePtr;
+using nnvm::NodeEntry;
+using nnvm::Graph;
+
+#define DEBUG_SUBGRAPH 0
+
+namespace sg { // sg stands for subgraph
+
+struct SimpleNode;
+using SimpleNodePtr = std::shared_ptr;
+
+/*!
+ * \brief Node of the undirected graph which replicates the network structures
+ * of the computational graph. It is used to ease the graph traversal for
finding
+ * subgraphs.
+ */
+struct SimpleNode {
+ static SimpleNodePtr Create() {
+return std::make_shared();
+ }
+ SimpleNode() : label(-1), node(nullptr) {}
+ /*! subgraph label */
+ int label;
+ /*! the original node in the computational graph it references*/
+ nnvm::Node* node;
+ /*!
+ * \brief output nodes of the current node
+ * key is node ptr and value is an array of indices standing for the entry
indices
+ * in key->inputs whose source is the current node.
+ */
+ std::unordered_map> outputs;
+}; // struct SimpleNode
+
+#if DEBUG_SUBGRAPH
+void PrintSubgraph(const std::vector& simple_nodes) {
+ std::string op_names = "";
+ for (size_t i = 0; i < simple_nodes.size(); ++i) {
+op_names += simple_nodes[i]->node->attrs.name + ' ';
+ }
+ LOG(INFO) << "Subgraph node names: " << op_names;
+}
+
+void PrintNodeEntry(const nnvm::NodeEntry& entry) {
+ std::string ret = "NodeEntry: node_name=" + entry.node->attrs.name
++ ", index=" + std::to_string(entry.index) + ", version=" +
std::to_string(entry.version);
+ LOG(INFO) << ret;
+}
+
+void PrintNodeEntries(const std::vector& entries) {
+ for (size_t i = 0; i < entries.size(); ++i) {
+PrintNodeEntry(*entries[i]);
+ }
+}
+#endif
+
+/*!
+ * \brief Given a MXNet computational graph, create an undirected graph from
it.
+ * \param g the MXNet computational graph
+ * \param simple_nodes the nodes of undirected graph in top sorted order
+ */
+void CreateSimpleGraph(const Graph& g,
+ std::vector* simple_nodes) {
+ const auto& indexed_graph = g.indexed_graph();
+ simple_nodes->reserve(indexed_graph.num_nodes());
+ DFSVisit(g.outputs, [&](const NodePtr& node) {
+SimpleNodePtr sn = SimpleNode::Create();
+sn->node = node.get();
+for (size_t i = 0; i < sn->node->inputs.size(); ++i) {
+ const auto& e = sn->node->inputs[i];
+ const auto input_nid = indexed_graph.node_id(e.node.get());
+ CHECK_LT(input_nid, simple_nodes->size());
+ auto& input_node_outputs = (*simple_nodes)[input_nid]->outputs;
+ auto it = input_node_outputs.find(sn->node);
+ if (it == input_node_outputs.end()) {
+input_node_outputs.emplace(sn->node, std::vector{i});
+ } else {
+it->second.push_back(i);
+ }
+}
+simple_nodes->emplace_back(std::move(sn));
+ });
+}
+
+/*!
+ * \brief Reset labels of the subgraph nodes to the original state
+ * and clear the vector of subgraph nodes.
+ */
+void ResetNodeLabels(const nnvm::Graph& g,
+ const std::vector& simple_nodes,
+ std::vector* subgraph_nodes) {
+ for (auto n : *subgraph_nodes) {
+const auto nid = g.indexed_graph().node_id(n);
+simple_nodes[nid]->label = -1;
+ }
+ subgraph_nodes->clear();
+}
+
+/*!
+ * \brief This function traverses the nodes in a computation graph from a
starting
+ * node following the input edges and output edges, and marks all nodes that
+ * can be accessed from the starting node. Before the function returns,
+ * it will conduct checking whether there is a loop between the potential
subgraph
+ * and the outside nodes. If so, add the node that should
[GitHub] reminisce commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
reminisce commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213503015
##
File path: src/operator/subgraph/partition_graph.cc
##
@@ -0,0 +1,774 @@
+/*
+ * 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.
+ */
+
+/*!
+ * Copyright (c) 2018 by Contributors
+ * \file partition_graph.cc
+ * \brief
+ */
+#include
+#include
+#include
+#include
+#include
+#include
+
+#include "./subgraph_property.h"
+
+namespace nnvm {
+NodePtr CreateVariableNode(const std::string& name);
+}
+
+namespace mxnet {
+
+namespace op {
+
+using nnvm::Symbol;
+using nnvm::Node;
+using nnvm::NodePtr;
+using nnvm::NodeEntry;
+using nnvm::Graph;
+
+#define DEBUG_SUBGRAPH 0
+
+namespace sg { // sg stands for subgraph
+
+struct SimpleNode;
+using SimpleNodePtr = std::shared_ptr;
+
+/*!
+ * \brief Node of the undirected graph which replicates the network structures
+ * of the computational graph. It is used to ease the graph traversal for
finding
+ * subgraphs.
+ */
+struct SimpleNode {
+ static SimpleNodePtr Create() {
+return std::make_shared();
+ }
+ SimpleNode() : label(-1), node(nullptr) {}
+ /*! subgraph label */
+ int label;
+ /*! the original node in the computational graph it references*/
+ nnvm::Node* node;
+ /*!
+ * \brief output nodes of the current node
+ * key is node ptr and value is an array of indices standing for the entry
indices
+ * in key->inputs whose source is the current node.
+ */
+ std::unordered_map> outputs;
+}; // struct SimpleNode
+
+#if DEBUG_SUBGRAPH
+void PrintSubgraph(const std::vector& simple_nodes) {
+ std::string op_names = "";
+ for (size_t i = 0; i < simple_nodes.size(); ++i) {
+op_names += simple_nodes[i]->node->attrs.name + ' ';
+ }
+ LOG(INFO) << "Subgraph node names: " << op_names;
+}
+
+void PrintNodeEntry(const nnvm::NodeEntry& entry) {
+ std::string ret = "NodeEntry: node_name=" + entry.node->attrs.name
++ ", index=" + std::to_string(entry.index) + ", version=" +
std::to_string(entry.version);
+ LOG(INFO) << ret;
+}
+
+void PrintNodeEntries(const std::vector& entries) {
+ for (size_t i = 0; i < entries.size(); ++i) {
+PrintNodeEntry(*entries[i]);
+ }
+}
+#endif
+
+/*!
+ * \brief Given a MXNet computational graph, create an undirected graph from
it.
+ * \param g the MXNet computational graph
+ * \param simple_nodes the nodes of undirected graph in top sorted order
+ */
+void CreateSimpleGraph(const Graph& g,
+ std::vector* simple_nodes) {
+ const auto& indexed_graph = g.indexed_graph();
+ simple_nodes->reserve(indexed_graph.num_nodes());
+ DFSVisit(g.outputs, [&](const NodePtr& node) {
+SimpleNodePtr sn = SimpleNode::Create();
+sn->node = node.get();
+for (size_t i = 0; i < sn->node->inputs.size(); ++i) {
+ const auto& e = sn->node->inputs[i];
+ const auto input_nid = indexed_graph.node_id(e.node.get());
+ CHECK_LT(input_nid, simple_nodes->size());
+ auto& input_node_outputs = (*simple_nodes)[input_nid]->outputs;
+ auto it = input_node_outputs.find(sn->node);
+ if (it == input_node_outputs.end()) {
+input_node_outputs.emplace(sn->node, std::vector{i});
+ } else {
+it->second.push_back(i);
+ }
+}
+simple_nodes->emplace_back(std::move(sn));
+ });
+}
+
+/*!
+ * \brief Reset labels of the subgraph nodes to the original state
+ * and clear the vector of subgraph nodes.
+ */
+void ResetNodeLabels(const nnvm::Graph& g,
+ const std::vector& simple_nodes,
+ std::vector* subgraph_nodes) {
+ for (auto n : *subgraph_nodes) {
+const auto nid = g.indexed_graph().node_id(n);
+simple_nodes[nid]->label = -1;
+ }
+ subgraph_nodes->clear();
+}
+
+/*!
+ * \brief This function traverses the nodes in a computation graph from a
starting
+ * node following the input edges and output edges, and marks all nodes that
+ * can be accessed from the starting node. Before the function returns,
+ * it will conduct checking whether there is a loop between the potential
subgraph
+ * and the outside nodes. If so, add the node that should
[GitHub] reminisce commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
reminisce commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213502884
##
File path: src/operator/subgraph/subgraph_property.h
##
@@ -0,0 +1,141 @@
+/*
+ * 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.
+ */
+
+#ifndef MXNET_OPERATOR_SUBGRAPH_SUBGRAPH_PROPERTY_H_
+#define MXNET_OPERATOR_SUBGRAPH_SUBGRAPH_PROPERTY_H_
+
+#include
+#include
+#include
+#include
+#include
+#include
+
+namespace mxnet {
+namespace op {
+
+/*
+ * This provides criteria for selecting nodes in a subgraph.
+ * When a node is passed to this object, the selection criteria may be changed.
+ * We can also specify what links we should use when traversing the neighbor
+ * nodes.
+ */
+class SubgraphSelector {
+ public:
+ virtual ~SubgraphSelector() {}
+ // Determine if the node should be selected for a subgraph.
+ virtual bool Select(const nnvm::Node ) = 0;
+ // Determine if the input node should be selected for a subgraph.
+ virtual bool SelectInput(const nnvm::Node , const nnvm::Node _node) =
0;
Review comment:
Done.
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] reminisce commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
reminisce commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213502853
##
File path: src/operator/subgraph/common.h
##
@@ -0,0 +1,237 @@
+/*
+ * 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.
+ */
+
+#ifndef MXNET_OPERATOR_SUBGRAPH_COMMON_H_
+#define MXNET_OPERATOR_SUBGRAPH_COMMON_H_
+
+#include
+#include
+#include
+#include "../elemwise_op_common.h"
+#include "../../executor/exec_pass.h"
+
+namespace mxnet {
+namespace op {
+
+inline uint32_t DefaultSubgraphOpNumInputs(const nnvm::NodeAttrs& attrs) {
+ const nnvm::Symbol& sym = *attrs.subgraphs[0];
+ return sym.ListInputNames(nnvm::Symbol::kAll).size();
+}
+
+inline uint32_t DefaultSubgraphOpNumOutputs(const nnvm::NodeAttrs& attrs) {
+ const nnvm::Symbol& sym = *attrs.subgraphs[0];
+ return sym.ListOutputNames().size();
+}
+
+inline std::vector DefaultSubgraphOpListInputs(const
nnvm::NodeAttrs& attrs) {
+ const nnvm::Symbol& sym = *attrs.subgraphs[0];
+ return sym.ListInputNames(nnvm::Symbol::kAll);
+}
+
+inline std::vector DefaultSubgraphOpListOutputs(const
nnvm::NodeAttrs& attrs) {
+ const nnvm::Symbol& sym = *attrs.subgraphs[0];
+ return sym.ListOutputNames();
+}
+
+inline bool DefaultSubgraphOpShape(const nnvm::NodeAttrs& attrs,
+ std::vector *in_shapes,
+ std::vector *out_shapes) {
+ using namespace exec;
+ const nnvm::Symbol& subgraph_sym = *attrs.subgraphs[0];
+ nnvm::Graph g;
+ g.outputs = subgraph_sym.outputs;
+ const auto& idx_g = g.indexed_graph();
+ CHECK_EQ(idx_g.input_nodes().size(), in_shapes->size());
+ CHECK_EQ(idx_g.outputs().size(), out_shapes->size());
+
+ // Put the input and output shapes to the shape vector.
+ nnvm::ShapeVector shapes(idx_g.num_node_entries());
+ const auto _nids = idx_g.input_nodes();
+ CHECK_EQ(input_nids.size(), in_shapes->size());
+ for (size_t i = 0; i < in_shapes->size(); i++) {
+auto eid = idx_g.entry_id(input_nids[i], 0);
+shapes[eid] = in_shapes->at(i);
+ }
+ CHECK_EQ(g.outputs.size(), out_shapes->size());
+ for (size_t i = 0; i < out_shapes->size(); i++) {
+auto eid = idx_g.entry_id(g.outputs[i]);
+shapes[eid] = out_shapes->at(i);
+ }
+
+ // Infer shape of the graph.
+ g.attrs["shape"] = std::make_shared(std::move(shapes));
+ g = exec::InferShape(std::move(g));
+
+ // Copy the inferred shape back to the input shapes and the output shapes.
+ shapes = g.GetAttr("shape");
+ // assign to in_shapes
+ for (size_t i = 0; i < in_shapes->size(); ++i) {
+const auto eid = idx_g.entry_id(input_nids[i], 0);
+SHAPE_ASSIGN_CHECK(*in_shapes, i, shapes[eid]);
+ }
+ // assign to out_shapes
+ for (size_t i = 0; i < g.outputs.size(); ++i) {
+const auto eid = idx_g.entry_id(g.outputs[i]);
+SHAPE_ASSIGN_CHECK(*out_shapes, i, shapes[eid]);
+ }
+ // Check if we have inferred the shapes correctly.
+ return g.GetAttr("shape_num_unknown_nodes") == 0;
+}
+
+inline bool DefaultSubgraphOpType(const nnvm::NodeAttrs& attrs,
+ std::vector *in_types,
+ std::vector *out_types) {
+ const nnvm::Symbol& subgraph_sym = *attrs.subgraphs[0];
+ nnvm::Graph g;
+ g.outputs = subgraph_sym.outputs;
+ const auto& idx_g = g.indexed_graph();
+ CHECK_EQ(idx_g.input_nodes().size(), in_types->size());
+ CHECK_EQ(idx_g.outputs().size(), out_types->size());
+
+ // Put the input and output data types to the dtype vector.
+ nnvm::DTypeVector types(idx_g.num_node_entries(), -1);
+ const auto _nids = idx_g.input_nodes();
+ CHECK_EQ(input_nids.size(), in_types->size());
+ for (size_t i = 0; i < in_types->size(); i++) {
+auto eid = idx_g.entry_id(input_nids[i], 0);
+types[eid] = in_types->at(i);
+ }
+ CHECK_EQ(g.outputs.size(), out_types->size());
+ for (size_t i = 0; i < out_types->size(); i++) {
+auto eid = idx_g.entry_id(g.outputs[i]);
+types[eid] = out_types->at(i);
+ }
+
+ // Infer data type of the graph.
+ g.attrs["dtype"] = std::make_shared(std::move(types));
+ g = exec::InferType(std::move(g));
+
+ types = g.GetAttr("dtype");
[GitHub] reminisce commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
reminisce commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213502485
##
File path: src/executor/graph_executor.cc
##
@@ -42,6 +43,7 @@ using namespace mxnet::common;
GraphExecutor::GraphExecutor() {
log_verbose_ = dmlc::GetEnv("MXNET_EXEC_VERBOSE_LOGGING", false);
need_grad_ = false;
+ subgraph_property_ = dmlc::GetEnv("MXNET_SUBGRAPH_BACKEND", std::string());
Review comment:
Right now, there is no real accelerator integrated with MXNet. Intel team is
going to submit their work on MKLDNN using this API set. We can add this evn
var along with their PR.
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] haojin2 commented on issue #12386: [MXNET-810] [WIP] Add support for more req patterns for bilinear sampler backward
haojin2 commented on issue #12386: [MXNET-810] [WIP] Add support for more req patterns for bilinear sampler backward URL: https://github.com/apache/incubator-mxnet/pull/12386#issuecomment-416768267 @eric-haibin-lin @reminisce @piiswrong @anirudh2290 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] haojin2 opened a new pull request #12386: [MXNET-810] [WIP] Add support for more req patterns for bilinear sampler backward
haojin2 opened a new pull request #12386: [MXNET-810] [WIP] Add support for more req patterns for bilinear sampler backward URL: https://github.com/apache/incubator-mxnet/pull/12386 ## Description ## Fix #8866. ## Checklist ## ### Essentials ### - [x] Changes are complete (i.e. I finished coding on this PR) - [x] 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) - [x] 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 - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [x] Support more req type patterns for backward of bilinear sampler - [ ] Corresponding unit tests ## Comments ## Currently test_bilinear_sampler is still marked as flaky, cannot un-leash the unit test for that, so blocked at this moment. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] metrofun edited a comment on issue #2811: where can i find rbm example by mxnet?
metrofun edited a comment on issue #2811: where can i find rbm example by mxnet? URL: https://github.com/apache/incubator-mxnet/issues/2811#issuecomment-416764491 The second post on RBM in MXNet was low on my priority, but here is the code I wrote for the second post https://gist.github.com/metrofun/5c76dc280b8ce19e56d673e19dc262c4 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] metrofun commented on issue #2811: where can i find rbm example by mxnet?
metrofun commented on issue #2811: where can i find rbm example by mxnet? URL: https://github.com/apache/incubator-mxnet/issues/2811#issuecomment-416764491 Second post on RBM in MXNet was low on my priority, but here is the code I wrote for the second post https://gist.github.com/metrofun/5c76dc280b8ce19e56d673e19dc262c4 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] vishaalkapoor commented on issue #2811: where can i find rbm example by mxnet?
vishaalkapoor commented on issue #2811: where can i find rbm example by mxnet? URL: https://github.com/apache/incubator-mxnet/issues/2811#issuecomment-416758822 @yzhliu Would you be able to close? Thank you! This is an automated message from the Apache Git Service. To respond to the message, please log on 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] terrytangyuan commented on issue #12162: Edit shape.array doc and some style improvements
terrytangyuan commented on issue #12162: Edit shape.array doc and some style
improvements
URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416758686
I believe that should be done in a separate PR. I doubt that `formatR` gives
the most readable code, e.g. I've rarely seen the following style but it
appears to be on the PR you referred to:
```
convolution_module <- function(net, kernel_size, pad_size, filter_count,
stride = c(1,
1), work_space = 2048, batch_norm = TRUE, down_pool = FALSE, up_pool =
FALSE,
```
or
```
if (Sys.getenv("R_GPU_ENABLE") != "" &
as.integer(Sys.getenv("R_GPU_ENABLE")) ==
1) {
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] vandanavk edited a comment on issue #12291: [MXNET-817] Fixes to speech recognition example
vandanavk edited a comment on issue #12291: [MXNET-817] Fixes to speech recognition example URL: https://github.com/apache/incubator-mxnet/pull/12291#issuecomment-416704408 @anirudhacharya Opened an issue https://github.com/apache/incubator-mxnet/issues/12384 for tracking the refactor This is an automated message from the Apache Git Service. To respond to the message, please log on 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] marcoabreu commented on issue #12379: Revert "Revert "Disable kvstore test (#11798)" (#12279)"
marcoabreu commented on issue #12379: Revert "Revert "Disable kvstore test (#11798)" (#12279)" URL: https://github.com/apache/incubator-mxnet/pull/12379#issuecomment-416752760 I can only merge if this PR passes CI This is an automated message from the Apache Git Service. To respond to the message, please log on 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] Roshrini commented on issue #10948: get stuck with subprocess in multithread
Roshrini commented on issue #10948: get stuck with subprocess in multithread URL: https://github.com/apache/incubator-mxnet/issues/10948#issuecomment-416740256 @HorsonLiu MXNet is not very thread-safe and so this is not supported yet. Adding it as a FeatureRequest. @sandeep-krishnamurthy Can you please tag this as [FeatureRequest, Backend] This is an automated message from the Apache Git Service. To respond to the message, please log on 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] vandanavk edited a comment on issue #12182: [MXNET-698] Remove Epoch training metric log
vandanavk edited a comment on issue #12182: [MXNET-698] Remove Epoch training metric log URL: https://github.com/apache/incubator-mxnet/pull/12182#issuecomment-416634452 @nswamy The following is the current behavior of the code. We need to reflect this in the "train-accuracy" log in base_module.py With Speedometer's `auto_reset=True` ``` INFO:root:Epoch[0] Batch [1-100] Speed: 45690.13 samples/sec accuracy=0.772123 INFO:root:Epoch[0] Batch [101-200] Speed: 50611.24 samples/sec accuracy=0.898594 . . . INFO:root:Epoch[0] Batch [801-900] Speed: 52047.39 samples/sec accuracy=0.950625 INFO:root:Epoch[0] Batch [901-938] Train-accuracy=0.944679 INFO:root:Epoch[0] Time cost=1.250 INFO:root:Epoch[0] Validation-accuracy=0.953125 ``` With `auto_reset=False` ``` INFO:root:Epoch[0] Batch [1-100] Speed: 16628.82 samples/sec accuracy=0.759746 INFO:root:Epoch[0] Batch [1-200] Speed: 40806.39 samples/sec accuracy=0.828980 . . . INFO:root:Epoch[0] Batch [1-900] Speed: 42613.81 samples/sec accuracy=0.911470 INFO:root:Epoch[0] Batch [1-938] Train-accuracy=0.912830 INFO:root:Epoch[0] Time cost=1.811 INFO:root:Epoch[0] Validation-accuracy=0.956509 ``` This is an automated message from the Apache Git Service. To respond to the message, please log on 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] haojin2 closed pull request #12306: SoftMin Operator
haojin2 closed pull request #12306: SoftMin Operator
URL: https://github.com/apache/incubator-mxnet/pull/12306
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/src/operator/contrib/ctc_loss-inl.h
b/src/operator/contrib/ctc_loss-inl.h
index 72209ae286c..9380be47451 100644
--- a/src/operator/contrib/ctc_loss-inl.h
+++ b/src/operator/contrib/ctc_loss-inl.h
@@ -409,7 +409,8 @@ class CTCLossOp : public Operator {
// since the input is activation before softmax and cudnn ctc takes softmax
// apply softmax to inputs first.
-mxnet_op::Softmax(s, data.dptr_, prob.dptr_,
data.shape_, 2, 1.0);
+mxnet_op::Softmax(
+ s, data.dptr_, prob.dptr_, data.shape_, 2, 1.0);
CUDNN_CALL(cudnnCTCLoss(s->dnn_handle_,
prob_desc_,
@@ -426,8 +427,8 @@ class CTCLossOp : public Operator {
workspace_bytes));
if (req_grad) {
- mxnet_op::SoftmaxGrad(s,
- prob.dptr_, grad.dptr_, grad.dptr_, data.shape_, 2, 1.0);
+ mxnet_op::SoftmaxGrad(
+s, prob.dptr_, grad.dptr_, grad.dptr_, data.shape_, 2, 1.0);
Assign(grad, mxnet::kWriteInplace, grad * alphabet_size);
}
}
diff --git a/src/operator/nn/softmax-inl.h b/src/operator/nn/softmax-inl.h
index 4a19db7c36b..c063e385f63 100644
--- a/src/operator/nn/softmax-inl.h
+++ b/src/operator/nn/softmax-inl.h
@@ -51,7 +51,7 @@ struct log_softmax_fwd {
};
-template
+template
inline void Softmax(Stream *s, DType *in, DType *out,
Shape shape, int axis, const DType temperature) {
index_t M = shape[axis];
@@ -65,30 +65,37 @@ inline void Softmax(Stream *s, DType *in, DType *out,
for (int i = 0; i < static_cast(N); ++i) {
index_t base = unravel_dot(i, sshape, stride);
-DType mmax = in[base];
+DType mmax = negate ? -in[base] : in[base];
+DType val;
for (index_t j = 1; j < M; ++j) {
- if (mmax < in[base + j*sa]) mmax = in[base + j*sa];
+ val = negate ? -in[base + j*sa] : in[base + j*sa];
+ if (mmax < val) mmax = val;
}
DType sum = DType(0);
+DType in_val;
// By default temperature is 1.0, and only in reinforcement training
// users would set it to other values.
// Adding a branch here to save the CPU 'divide-by-1' computation at
runtime
if (temperature == 1.0) {
for (index_t j = 0; j < M; ++j) {
-sum += std::exp(in[base + j*sa] - mmax);
+in_val = negate ? -in[base + j*sa] : in[base + j*sa];
+sum += std::exp(in_val - mmax);
}
for (index_t j = 0; j < M; ++j) {
-out[base + j*sa] = OP::Map(in[base + j*sa] - mmax, sum);
+in_val = negate ? -in[base + j*sa] : in[base + j*sa];
+out[base + j*sa] = OP::Map(in_val - mmax, sum);
}
} else {
for (index_t j = 0; j < M; ++j) {
-sum += std::exp((in[base + j*sa] - mmax)/temperature);
+in_val = negate ? -in[base + j*sa] : in[base + j*sa];
+sum += std::exp((in_val - mmax)/temperature);
}
for (index_t j = 0; j < M; ++j) {
-out[base + j*sa] = OP::Map((in[base + j*sa] - mmax)/temperature, sum);
+in_val = negate ? -in[base + j*sa] : in[base + j*sa];
+out[base + j*sa] = OP::Map((in_val - mmax)/temperature, sum);
}
}
}
@@ -111,7 +118,7 @@ struct log_softmax_bwd {
};
-template
+template
inline void SoftmaxGrad(Stream *s, DType *out, DType *ograd,
DType *igrad, Shape shape, int axis,
const DType temperature) {
@@ -137,12 +144,16 @@ inline void SoftmaxGrad(Stream *s, DType *out, DType
*ograd,
DType final_result;
if (temperature == 1.0) {
for (index_t j = 0; j < M; ++j) {
-final_result = OP2::Map(ograd[base + j*sa], out[base + j*sa], sum);
+final_result = negate ?
+ -OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) :
+ OP2::Map(ograd[base + j*sa], out[base + j*sa], sum);
KERNEL_ASSIGN(igrad[base + j*sa], Req, final_result);
}
} else {
for (index_t j = 0; j < M; ++j) {
-final_result = OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) /
temperature;
+final_result = negate ?
+ -OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) /
temperature :
+ OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) /
temperature;
KERNEL_ASSIGN(igrad[base + j*sa], Req, final_result);
}
}
@@ -151,7 +162,7 @@ inline void SoftmaxGrad(Stream *s, DType *out, DType
*ograd,
#ifdef __CUDACC__
-template
+template
__global__ void softmax_compute_kernel(DType *in, DType *out, index_t M, int
axis,
[GitHub] haojin2 opened a new pull request #12306: SoftMin Operator
haojin2 opened a new pull request #12306: SoftMin Operator URL: https://github.com/apache/incubator-mxnet/pull/12306 ## Description ## Support softmin function: https://pytorch.org/docs/master/_modules/torch/nn/functional.html#softmin ## Checklist ## ### Essentials ### - [x] Changes are complete (i.e. I finished coding on this PR) - [x] 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) - [x] 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 - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [x] Softmin operator - [x] Corresponding unit tests ## Comments ## Passed more than 1 trials on both CPU & GPU: ``` MXNET_TEST_COUNT=1 nosetests -s --verbose tests/python/unittest/test_operator.py:test_softmin [INFO] Setting module np/mx/python random seeds, use MXNET_MODULE_SEED=53121875 to reproduce. test_operator.test_softmin ... ok -- Ran 1 test in 670.955s OK ``` ``` MXNET_TEST_COUNT=1 nosetests -s --verbose tests/python/gpu/test_operator_gpu.py:test_softmin [INFO] Setting module np/mx/python random seeds, use MXNET_MODULE_SEED=207344053 to reproduce. test_operator_gpu.test_softmin ... ok -- Ran 1 test in 946.723s OK ``` This is an automated message from the Apache Git Service. To respond to the message, please log on 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] haojin2 commented on issue #12306: SoftMin Operator
haojin2 commented on issue #12306: SoftMin Operator URL: https://github.com/apache/incubator-mxnet/pull/12306#issuecomment-416730180 @szha This is an automated message from the Apache Git Service. To respond to the message, please log on 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] anirudhacharya commented on issue #12162: Edit shape.array doc and some style improvements
anirudhacharya commented on issue #12162: Edit shape.array doc and some style improvements URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416723075 @terrytangyuan Because we are in the process of having uniform standards for the R-package and eventually have lintR run on the pipeline. For example - https://github.com/apache/incubator-mxnet/pull/12360 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] terrytangyuan commented on issue #12162: Edit shape.array doc and some style improvements
terrytangyuan commented on issue #12162: Edit shape.array doc and some style improvements URL: https://github.com/apache/incubator-mxnet/pull/12162#issuecomment-416712675 What's the difference? I am just making some enhancements myself manually for this particular file. All the changes are valid here. Others feel free to use formatR themselves for enhancement for the overall package. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] mseth10 opened a new pull request #12385: fixed flaky test issue for test_operator_gpu.test_convolution_grouping
mseth10 opened a new pull request #12385: fixed flaky test issue for test_operator_gpu.test_convolution_grouping URL: https://github.com/apache/incubator-mxnet/pull/12385 ## Description ## Issue not reproducible on Ubuntu. Tolerance parameter (atol) relaxed. This should fix the flakiness issue on Windows #12219 since mismatch percentage is small. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [x] Changes are complete (i.e. I finished coding on this PR) - [x] 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) - [x] 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 - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [x] Tolerance parameter atol modified to 1e-3 ## Comments ## - Passed more than 10,000 times on GPU - @haojin2 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] marcoabreu closed pull request #12381: A solution to prevent zombie containers locally and in CI
marcoabreu closed pull request #12381: A solution to prevent zombie containers
locally and in CI
URL: https://github.com/apache/incubator-mxnet/pull/12381
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/ci/README.md b/ci/README.md
index 548e9cb9b04..69308756943 100644
--- a/ci/README.md
+++ b/ci/README.md
@@ -59,6 +59,20 @@ To work inside a container with a shell you can do:
When building, the artifacts are located in the build/ directory in the
project root. In case
`build.py -a` is invoked, the artifacts are located in build./
+# Docker container cleanup (Zombie containers)
+Docker has a client-server architecture, so when the program that is executing
the docker client
+dies or receieves a signal, the container keeps running as it's started by the
docker daemon.
+We implement signal handlers that catch sigterm and sigint and cleanup
containers before exit. In
+Jenkins there's not enough time between sigterm and sigkill so we guarantee
that containers are not
+left running by propagating environment variables used by the Jenkins process
tree killer to
+identify which process to kill when the job is stopped. This has the effect of
stopping the
+container given that the process inside the container is terminated.
+
+How to test this is working propperly: On the console you can hit ^C while a
container is running
+(not just building) and see that the container is stopped by running `docker
ps` on another
+terminal. In Jenkins this has been tested by stopping the job which has
containers running and
+verifying that the container stops shortly afterwards by running docker ps.
+
## Add a platform
To add a platform, you should add the appropriate dockerfile in
diff --git a/ci/build.py b/ci/build.py
index f1a5e99e2d0..df9e97bdb5f 100755
--- a/ci/build.py
+++ b/ci/build.py
@@ -23,26 +23,67 @@
"""
__author__ = 'Marco de Abreu, Kellen Sunderland, Anton Chernov, Pedro Larroy'
-__version__ = '0.2'
+__version__ = '0.3'
import argparse
import glob
import logging
+import os
import re
import shutil
import subprocess
import sys
import tempfile
-from copy import deepcopy
from itertools import chain
-from subprocess import call, check_call, check_output
+from subprocess import check_call, check_output
from typing import *
from util import *
+import docker
+import docker.models
+import docker.errors
+import signal
+import atexit
import pprint
-import requests
-CCACHE_MAXSIZE = '500G'
+class Cleanup:
+"""A class to cleanup containers"""
+def __init__(self):
+self.containers = set()
+self.docker_stop_timeout = 3
+
+def add_container(self, container: docker.models.containers.Container):
+assert isinstance(container, docker.models.containers.Container)
+self.containers.add(container)
+
+def remove_container(self, container: docker.models.containers.Container):
+assert isinstance(container, docker.models.containers.Container)
+self.containers.remove(container)
+
+def _cleanup_containers(self):
+if self.containers:
+logging.warning("Cleaning up containers")
+else:
+return
+# noinspection PyBroadException
+try:
+stop_timeout = int(os.environ.get("DOCKER_STOP_TIMEOUT",
self.docker_stop_timeout))
+except Exception:
+stop_timeout = 3
+for container in self.containers:
+try:
+container.stop(timeout=stop_timeout)
+logging.info("☠: stopped container %s",
trim_container_id(container.id))
+container.remove()
+logging.info(": removed container %s",
trim_container_id(container.id))
+except Exception as e:
+logging.exception(e)
+self.containers.clear()
+logging.info("Cleaning up containers finished.")
+
+def __call__(self):
+"""Perform cleanup"""
+self._cleanup_containers()
def get_dockerfiles_path():
@@ -115,7 +156,10 @@ def run_cmd():
run_cmd()
# Get image id by reading the tag. It's guaranteed (except race condition)
that the tag exists. Otherwise, the
# check_call would have failed
-return _get_local_image_id(docker_binary=docker_binary, docker_tag=tag)
+image_id = _get_local_image_id(docker_binary=docker_binary, docker_tag=tag)
+if not image_id:
+raise FileNotFoundError('Unable to find docker image id matching with
{}'.format(tag))
+return image_id
def _get_local_image_id(docker_binary, docker_tag):
@@ -137,10 +181,11 @@ def buildir() -> str:
def default_ccache_dir() -> str:
+""":return: ccache directory for the current platform"""
# Share ccache across containers
if 'CCACHE_DIR' in
[incubator-mxnet] branch master updated: A solution to prevent zombie containers locally and in CI (#12381)
This is an automated email from the ASF dual-hosted git repository.
marcoabreu 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 e2a3eef A solution to prevent zombie containers locally and in CI
(#12381)
e2a3eef is described below
commit e2a3eef349cb6643c08a7840d8cbd43b38fedfd5
Author: Pedro Larroy <928489+lar...@users.noreply.github.com>
AuthorDate: Tue Aug 28 21:16:31 2018 +0200
A solution to prevent zombie containers locally and in CI (#12381)
Fix pylint, mypy, and pycharm code inspection warnings
---
ci/README.md | 14 +++
ci/build.py | 304 ---
2 files changed, 243 insertions(+), 75 deletions(-)
diff --git a/ci/README.md b/ci/README.md
index 548e9cb..6930875 100644
--- a/ci/README.md
+++ b/ci/README.md
@@ -59,6 +59,20 @@ To work inside a container with a shell you can do:
When building, the artifacts are located in the build/ directory in the
project root. In case
`build.py -a` is invoked, the artifacts are located in build./
+# Docker container cleanup (Zombie containers)
+Docker has a client-server architecture, so when the program that is executing
the docker client
+dies or receieves a signal, the container keeps running as it's started by the
docker daemon.
+We implement signal handlers that catch sigterm and sigint and cleanup
containers before exit. In
+Jenkins there's not enough time between sigterm and sigkill so we guarantee
that containers are not
+left running by propagating environment variables used by the Jenkins process
tree killer to
+identify which process to kill when the job is stopped. This has the effect of
stopping the
+container given that the process inside the container is terminated.
+
+How to test this is working propperly: On the console you can hit ^C while a
container is running
+(not just building) and see that the container is stopped by running `docker
ps` on another
+terminal. In Jenkins this has been tested by stopping the job which has
containers running and
+verifying that the container stops shortly afterwards by running docker ps.
+
## Add a platform
To add a platform, you should add the appropriate dockerfile in
diff --git a/ci/build.py b/ci/build.py
index f1a5e99..df9e97b 100755
--- a/ci/build.py
+++ b/ci/build.py
@@ -23,26 +23,67 @@
"""
__author__ = 'Marco de Abreu, Kellen Sunderland, Anton Chernov, Pedro Larroy'
-__version__ = '0.2'
+__version__ = '0.3'
import argparse
import glob
import logging
+import os
import re
import shutil
import subprocess
import sys
import tempfile
-from copy import deepcopy
from itertools import chain
-from subprocess import call, check_call, check_output
+from subprocess import check_call, check_output
from typing import *
from util import *
+import docker
+import docker.models
+import docker.errors
+import signal
+import atexit
import pprint
-import requests
-CCACHE_MAXSIZE = '500G'
+class Cleanup:
+"""A class to cleanup containers"""
+def __init__(self):
+self.containers = set()
+self.docker_stop_timeout = 3
+
+def add_container(self, container: docker.models.containers.Container):
+assert isinstance(container, docker.models.containers.Container)
+self.containers.add(container)
+
+def remove_container(self, container: docker.models.containers.Container):
+assert isinstance(container, docker.models.containers.Container)
+self.containers.remove(container)
+
+def _cleanup_containers(self):
+if self.containers:
+logging.warning("Cleaning up containers")
+else:
+return
+# noinspection PyBroadException
+try:
+stop_timeout = int(os.environ.get("DOCKER_STOP_TIMEOUT",
self.docker_stop_timeout))
+except Exception:
+stop_timeout = 3
+for container in self.containers:
+try:
+container.stop(timeout=stop_timeout)
+logging.info("☠: stopped container %s",
trim_container_id(container.id))
+container.remove()
+logging.info(": removed container %s",
trim_container_id(container.id))
+except Exception as e:
+logging.exception(e)
+self.containers.clear()
+logging.info("Cleaning up containers finished.")
+
+def __call__(self):
+"""Perform cleanup"""
+self._cleanup_containers()
def get_dockerfiles_path():
@@ -115,7 +156,10 @@ def build_docker(platform: str, docker_binary: str,
registry: str, num_retries:
run_cmd()
# Get image id by reading the tag. It's guaranteed (except race condition)
that the tag exists. Otherwise, the
# check_call would have failed
-return _get_local_image_id(docker_binary=docker_binary, docker_tag=tag)
+image_id =
[GitHub] zheng-da commented on issue #12269: get memory error when running a model exported from gluon model zoo
zheng-da commented on issue #12269: get memory error when running a model exported from gluon model zoo URL: https://github.com/apache/incubator-mxnet/issues/12269#issuecomment-416707002 Thanks. It seems mxnet should report an error instead of failing with a segfault. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] zheng-da closed issue #12269: get memory error when running a model exported from gluon model zoo
zheng-da closed issue #12269: get memory error when running a model exported from gluon model zoo URL: https://github.com/apache/incubator-mxnet/issues/12269 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] vandanavk commented on issue #12291: [MXNET-817] Fixes to speech recognition example
vandanavk commented on issue #12291: [MXNET-817] Fixes to speech recognition example URL: https://github.com/apache/incubator-mxnet/pull/12291#issuecomment-416704408 Opened an issue https://github.com/apache/incubator-mxnet/issues/12384 for tracking the refactor This is an automated message from the Apache Git Service. To respond to the message, please log on 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] vandanavk opened a new issue #12384: Refactor speech recognition example
vandanavk opened a new issue #12384: Refactor speech recognition example URL: https://github.com/apache/incubator-mxnet/issues/12384 ## Description - Cleanup/simplify speech recognition example - Refactor singleton usage - Fix Python3 failures https://github.com/apache/incubator-mxnet/issues/11042 Package used (Python/R/Scala/Julia): Python 2 and 3 ## Build info (Required if built from source) Compiler (gcc/clang/mingw/visual studio): MXNet commit hash: 6a7bfe905bafe94a33eccd0cb8bc42f0d667f606 ## Related issues https://github.com/apache/incubator-mxnet/issues/12024 PR https://github.com/apache/incubator-mxnet/pull/12291 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] Roshrini commented on issue #12364: Importing PyTorch when using ONNX causes a segmentation fault
Roshrini commented on issue #12364: Importing PyTorch when using ONNX causes a segmentation fault URL: https://github.com/apache/incubator-mxnet/issues/12364#issuecomment-416702032 @DatCorno Thanks for verfying this. @sandeep-krishnamurthy Can you please close this issue? This is an automated message from the Apache Git Service. To respond to the message, please log on 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. zhasheng 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 cde9c38 Bump the publish timestamp. cde9c38 is described below commit cde9c38fb2c6d9d6f0c6e37dfa0956e7adb1b136 Author: mxnet-ci AuthorDate: Tue Aug 28 18:56:03 2018 + 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..30d59a1 --- /dev/null +++ b/date.txt @@ -0,0 +1 @@ +Tue Aug 28 18:56:03 UTC 2018
[GitHub] anirudh2290 commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
anirudh2290 commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213432161
##
File path: src/operator/subgraph/partition_graph.cc
##
@@ -0,0 +1,774 @@
+/*
+ * 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.
+ */
+
+/*!
+ * Copyright (c) 2018 by Contributors
+ * \file partition_graph.cc
+ * \brief
+ */
+#include
+#include
+#include
+#include
+#include
+#include
+
+#include "./subgraph_property.h"
+
+namespace nnvm {
+NodePtr CreateVariableNode(const std::string& name);
+}
+
+namespace mxnet {
+
+namespace op {
+
+using nnvm::Symbol;
+using nnvm::Node;
+using nnvm::NodePtr;
+using nnvm::NodeEntry;
+using nnvm::Graph;
+
+#define DEBUG_SUBGRAPH 0
+
+namespace sg { // sg stands for subgraph
+
+struct SimpleNode;
+using SimpleNodePtr = std::shared_ptr;
+
+/*!
+ * \brief Node of the undirected graph which replicates the network structures
+ * of the computational graph. It is used to ease the graph traversal for
finding
+ * subgraphs.
+ */
+struct SimpleNode {
+ static SimpleNodePtr Create() {
+return std::make_shared();
+ }
+ SimpleNode() : label(-1), node(nullptr) {}
+ /*! subgraph label */
+ int label;
+ /*! the original node in the computational graph it references*/
+ nnvm::Node* node;
+ /*!
+ * \brief output nodes of the current node
+ * key is node ptr and value is an array of indices standing for the entry
indices
+ * in key->inputs whose source is the current node.
+ */
+ std::unordered_map> outputs;
+}; // struct SimpleNode
+
+#if DEBUG_SUBGRAPH
+void PrintSubgraph(const std::vector& simple_nodes) {
+ std::string op_names = "";
+ for (size_t i = 0; i < simple_nodes.size(); ++i) {
+op_names += simple_nodes[i]->node->attrs.name + ' ';
+ }
+ LOG(INFO) << "Subgraph node names: " << op_names;
+}
+
+void PrintNodeEntry(const nnvm::NodeEntry& entry) {
+ std::string ret = "NodeEntry: node_name=" + entry.node->attrs.name
++ ", index=" + std::to_string(entry.index) + ", version=" +
std::to_string(entry.version);
+ LOG(INFO) << ret;
+}
+
+void PrintNodeEntries(const std::vector& entries) {
+ for (size_t i = 0; i < entries.size(); ++i) {
+PrintNodeEntry(*entries[i]);
+ }
+}
+#endif
+
+/*!
+ * \brief Given a MXNet computational graph, create an undirected graph from
it.
+ * \param g the MXNet computational graph
+ * \param simple_nodes the nodes of undirected graph in top sorted order
+ */
+void CreateSimpleGraph(const Graph& g,
+ std::vector* simple_nodes) {
+ const auto& indexed_graph = g.indexed_graph();
+ simple_nodes->reserve(indexed_graph.num_nodes());
+ DFSVisit(g.outputs, [&](const NodePtr& node) {
+SimpleNodePtr sn = SimpleNode::Create();
+sn->node = node.get();
+for (size_t i = 0; i < sn->node->inputs.size(); ++i) {
+ const auto& e = sn->node->inputs[i];
+ const auto input_nid = indexed_graph.node_id(e.node.get());
+ CHECK_LT(input_nid, simple_nodes->size());
+ auto& input_node_outputs = (*simple_nodes)[input_nid]->outputs;
+ auto it = input_node_outputs.find(sn->node);
+ if (it == input_node_outputs.end()) {
+input_node_outputs.emplace(sn->node, std::vector{i});
+ } else {
+it->second.push_back(i);
+ }
+}
+simple_nodes->emplace_back(std::move(sn));
+ });
+}
+
+/*!
+ * \brief Reset labels of the subgraph nodes to the original state
+ * and clear the vector of subgraph nodes.
+ */
+void ResetNodeLabels(const nnvm::Graph& g,
+ const std::vector& simple_nodes,
+ std::vector* subgraph_nodes) {
+ for (auto n : *subgraph_nodes) {
+const auto nid = g.indexed_graph().node_id(n);
+simple_nodes[nid]->label = -1;
+ }
+ subgraph_nodes->clear();
+}
+
+/*!
+ * \brief This function traverses the nodes in a computation graph from a
starting
+ * node following the input edges and output edges, and marks all nodes that
+ * can be accessed from the starting node. Before the function returns,
+ * it will conduct checking whether there is a loop between the potential
subgraph
+ * and the outside nodes. If so, add the node that
[GitHub] anirudh2290 commented on a change in pull request #12157: Subgraph API for integrating accelerators with MXNet
anirudh2290 commented on a change in pull request #12157: Subgraph API for
integrating accelerators with MXNet
URL: https://github.com/apache/incubator-mxnet/pull/12157#discussion_r213432803
##
File path: src/operator/subgraph/partition_graph.cc
##
@@ -0,0 +1,774 @@
+/*
+ * 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.
+ */
+
+/*!
+ * Copyright (c) 2018 by Contributors
+ * \file partition_graph.cc
+ * \brief
+ */
+#include
+#include
+#include
+#include
+#include
+#include
+
+#include "./subgraph_property.h"
+
+namespace nnvm {
+NodePtr CreateVariableNode(const std::string& name);
+}
+
+namespace mxnet {
+
+namespace op {
+
+using nnvm::Symbol;
+using nnvm::Node;
+using nnvm::NodePtr;
+using nnvm::NodeEntry;
+using nnvm::Graph;
+
+#define DEBUG_SUBGRAPH 0
+
+namespace sg { // sg stands for subgraph
+
+struct SimpleNode;
+using SimpleNodePtr = std::shared_ptr;
+
+/*!
+ * \brief Node of the undirected graph which replicates the network structures
+ * of the computational graph. It is used to ease the graph traversal for
finding
+ * subgraphs.
+ */
+struct SimpleNode {
+ static SimpleNodePtr Create() {
+return std::make_shared();
+ }
+ SimpleNode() : label(-1), node(nullptr) {}
+ /*! subgraph label */
+ int label;
+ /*! the original node in the computational graph it references*/
+ nnvm::Node* node;
+ /*!
+ * \brief output nodes of the current node
+ * key is node ptr and value is an array of indices standing for the entry
indices
+ * in key->inputs whose source is the current node.
+ */
+ std::unordered_map> outputs;
+}; // struct SimpleNode
+
+#if DEBUG_SUBGRAPH
+void PrintSubgraph(const std::vector& simple_nodes) {
+ std::string op_names = "";
+ for (size_t i = 0; i < simple_nodes.size(); ++i) {
+op_names += simple_nodes[i]->node->attrs.name + ' ';
+ }
+ LOG(INFO) << "Subgraph node names: " << op_names;
+}
+
+void PrintNodeEntry(const nnvm::NodeEntry& entry) {
+ std::string ret = "NodeEntry: node_name=" + entry.node->attrs.name
++ ", index=" + std::to_string(entry.index) + ", version=" +
std::to_string(entry.version);
+ LOG(INFO) << ret;
+}
+
+void PrintNodeEntries(const std::vector& entries) {
+ for (size_t i = 0; i < entries.size(); ++i) {
+PrintNodeEntry(*entries[i]);
+ }
+}
+#endif
+
+/*!
+ * \brief Given a MXNet computational graph, create an undirected graph from
it.
+ * \param g the MXNet computational graph
+ * \param simple_nodes the nodes of undirected graph in top sorted order
+ */
+void CreateSimpleGraph(const Graph& g,
+ std::vector* simple_nodes) {
+ const auto& indexed_graph = g.indexed_graph();
+ simple_nodes->reserve(indexed_graph.num_nodes());
+ DFSVisit(g.outputs, [&](const NodePtr& node) {
+SimpleNodePtr sn = SimpleNode::Create();
+sn->node = node.get();
+for (size_t i = 0; i < sn->node->inputs.size(); ++i) {
+ const auto& e = sn->node->inputs[i];
+ const auto input_nid = indexed_graph.node_id(e.node.get());
+ CHECK_LT(input_nid, simple_nodes->size());
+ auto& input_node_outputs = (*simple_nodes)[input_nid]->outputs;
+ auto it = input_node_outputs.find(sn->node);
+ if (it == input_node_outputs.end()) {
+input_node_outputs.emplace(sn->node, std::vector{i});
+ } else {
+it->second.push_back(i);
+ }
+}
+simple_nodes->emplace_back(std::move(sn));
+ });
+}
+
+/*!
+ * \brief Reset labels of the subgraph nodes to the original state
+ * and clear the vector of subgraph nodes.
+ */
+void ResetNodeLabels(const nnvm::Graph& g,
+ const std::vector& simple_nodes,
+ std::vector* subgraph_nodes) {
+ for (auto n : *subgraph_nodes) {
+const auto nid = g.indexed_graph().node_id(n);
+simple_nodes[nid]->label = -1;
+ }
+ subgraph_nodes->clear();
+}
+
+/*!
+ * \brief This function traverses the nodes in a computation graph from a
starting
+ * node following the input edges and output edges, and marks all nodes that
+ * can be accessed from the starting node. Before the function returns,
+ * it will conduct checking whether there is a loop between the potential
subgraph
+ * and the outside nodes. If so, add the node that
[GitHub] yzhliu closed issue #9745: MultiIter example should also return pad in next(self)
yzhliu closed issue #9745: MultiIter example should also return pad in next(self) URL: https://github.com/apache/incubator-mxnet/issues/9745 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] anirudhacharya commented on issue #12291: [MXNET-817] Fixes to speech recognition example
anirudhacharya commented on issue #12291: [MXNET-817] Fixes to speech recognition example URL: https://github.com/apache/incubator-mxnet/pull/12291#issuecomment-416700409 LGTM for most parts. Is the singleton refactor being tracked anywhere? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] vandanavk commented on issue #12291: [MXNET-817] Fixes to speech recognition example
vandanavk commented on issue #12291: [MXNET-817] Fixes to speech recognition example URL: https://github.com/apache/incubator-mxnet/pull/12291#issuecomment-416697526 Is this PR good to go? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] sad- commented on issue #9745: MultiIter example should also return pad in next(self)
sad- commented on issue #9745: MultiIter example should also return pad in next(self) URL: https://github.com/apache/incubator-mxnet/issues/9745#issuecomment-416692175 @yizhi can we close this out? doc example has been fixed/made more explicit on how to write a custom DataIter here: https://mxnet.incubator.apache.org/tutorials/basic/data.html#custom-iterator This is an automated message from the Apache Git Service. To respond to the message, please log on 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] StephanieYuan commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
StephanieYuan commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213422213 ## File path: contrib/svrg_optimization_python/src/svrg_optimizer.py ## @@ -0,0 +1,131 @@ +# 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. +"""A `SVRGOptimizer` encapsulates two optimizers to accommodate SVRG optimization logic. +""" + + +import mxnet as mx + + +@mx.optimizer.register +class AssignmentOptimizer(mx.optimizer.Optimizer): +def update(self, index, weight, grad, state): +weight[:] = grad + +@mx.optimizer.register +class SVRGOptimizer(mx.optimizer.Optimizer): +"""SVRGOptimizer is a wrapper class for two optimizers: one for accumulating full gradients and the other Review comment: For the previous benchmarks SVRG is used with SGD. Meanwhile I did some exploration of using NAG + SVRG and Adam + SVRG. I think it will be valuable to benchmark those optimizers with SVRG too. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] StephanieYuan commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
StephanieYuan commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213422213 ## File path: contrib/svrg_optimization_python/src/svrg_optimizer.py ## @@ -0,0 +1,131 @@ +# 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. +"""A `SVRGOptimizer` encapsulates two optimizers to accommodate SVRG optimization logic. +""" + + +import mxnet as mx + + +@mx.optimizer.register +class AssignmentOptimizer(mx.optimizer.Optimizer): +def update(self, index, weight, grad, state): +weight[:] = grad + +@mx.optimizer.register +class SVRGOptimizer(mx.optimizer.Optimizer): +"""SVRGOptimizer is a wrapper class for two optimizers: one for accumulating full gradients and the other Review comment: For the previous benchmarks SVRG is used with SGD, I did some exploration of using NAG + SVRG and Adam + SVRG. I think it will be valuable to benchmark those optimizers with SVRG too. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213415709
##
File path: contrib/svrg_optimization_python/test_svrg_train.py
##
@@ -0,0 +1,84 @@
+# 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 mxnet as mx
+import numpy as np
+from src.svrg_module import SVRGModule
+
+
+def test_svrg_intermediate_level_api(num_epoch):
+"""Test intermediate level svrgmodule API where the training process
+need to be explicitly defined. KVstore is not explicitly created.
+"""
+di, mod = create_network()
+mod.bind(data_shapes=di.provide_data, label_shapes=di.provide_label)
+mod.init_params(initializer=mx.init.Uniform(0.01), allow_missing=False,
force_init=False, allow_extra=False)
+kv = mx.kv.create("local")
+mod.init_optimizer(kvstore=kv, optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),))
+metrics = mx.metric.create("mse")
+for e in range(num_epoch):
+metrics.reset()
+if e % (mod.update_freq) == 0:
+mod.update_full_grads(di)
+di.reset()
+for batch in di:
+mod.forward_backward(data_batch=batch)
+mod.update()
+mod.update_metric(metrics, batch.label)
+print('Epoch[%d] Time cost=%.3f', e, metrics.get())
Review comment:
Please use logging.
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213412718
##
File path: contrib/svrg_optimization_python/src/svrg_module.py
##
@@ -0,0 +1,581 @@
+# 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.
+"""A `SVRGModule` implements the `Module` API by wrapping an auxiliary module
to perform
+SVRG optimization logic.
+"""
+
+import mxnet as mx
+import time
+import logging
+from svrg_optimizer import SVRGOptimizer
+from mxnet.module import Module
+
+
+class SVRGModule(Module):
+"""SVRGModule is a module that encapsulates two Modules to accommodate the
SVRG optimization technique.
+It is functionally the same as Module API, except it is implemented using
SVRG optimization logic.
+
+Parameters
+--
+symbol : Symbol
+data_names : list of str
+Defaults to `('data')` for a typical model used in image
classification.
+label_names : list of str
+Defaults to `('softmax_label')` for a typical model used in image
+classification.
+logger : Logger
+Defaults to `logging`.
+context : Context or list of Context
+Defaults to ``mx.cpu()``.
+work_load_list : list of number
+Default ``None``, indicating uniform workload.
+fixed_param_names: list of str
+Default ``None``, indicating no network parameters are fixed.
+state_names : list of str
+states are similar to data and label, but not provided by data
iterator.
+Instead they are initialized to 0 and can be set by `set_states()`.
+group2ctxs : dict of str to context or list of context,
+ or list of dict of str to context
+Default is `None`. Mapping the `ctx_group` attribute to the context
assignment.
+compression_params : dict
+Specifies type of gradient compression and additional arguments
depending
+on the type of compression being used. For example, 2bit compression
requires a threshold.
+Arguments would then be {'type':'2bit', 'threshold':0.5}
+See mxnet.KVStore.set_gradient_compression method for more details on
gradient compression.
+update_freq: int
+Specifies the number of times to update the full gradients to be used
in the SVRG optimization. For instance,
+update_freq = 2 will calculates the gradients over all data every two
epochs
+Examples
+
+>>> # An example of declaring and using SVRGModule.
+>>> mod = mod = SVRGModule(symbol=lro, data_names=['data'],
label_names=['lin_reg_label'], update_freq=2)
+>>> mod.fit(di, eval_metric='mse', optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),),
+>>> num_epoch=num_epoch, kvstore='local')
+"""
+
+def __init__(self, symbol, data_names=('data',),
label_names=('softmax_label',),
+ logger=logging, context=mx.cpu(), work_load_list=None,
+ fixed_param_names=None, state_names=None, group2ctxs=None,
+ compression_params=None, update_freq=None):
+super(SVRGModule, self).__init__(symbol, data_names=data_names,
label_names=label_names, logger=logger,
+ context=context,
work_load_list=work_load_list,
+ fixed_param_names=fixed_param_names,
state_names=state_names,
+ group2ctxs=group2ctxs,
compression_params=compression_params)
+
+# Type check update_frequency
+if isinstance(update_freq, int):
+self.update_freq = update_freq
+else:
+raise TypeError("update_freq must be an integer")
+
+self._mod_aux = mx.mod.Module(symbol, data_names, label_names, logger,
context, work_load_list,
+ fixed_param_names, state_names,
group2ctxs, compression_params)
+
+self._param_dict = [{} for ctx in self._context]
+
+def _reset_bind(self):
+"""Internal function to reset binded state."""
+super(SVRGModule, self)._reset_bind()
+self._mod_aux._reset_bind()
+
+
+def reshape(self,
[GitHub] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213406997
##
File path: contrib/svrg_optimization_python/src/svrg_module.py
##
@@ -0,0 +1,581 @@
+# 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.
+"""A `SVRGModule` implements the `Module` API by wrapping an auxiliary module
to perform
+SVRG optimization logic.
+"""
+
+import mxnet as mx
+import time
+import logging
+from svrg_optimizer import SVRGOptimizer
+from mxnet.module import Module
+
+
+class SVRGModule(Module):
+"""SVRGModule is a module that encapsulates two Modules to accommodate the
SVRG optimization technique.
+It is functionally the same as Module API, except it is implemented using
SVRG optimization logic.
+
+Parameters
+--
+symbol : Symbol
+data_names : list of str
+Defaults to `('data')` for a typical model used in image
classification.
+label_names : list of str
+Defaults to `('softmax_label')` for a typical model used in image
+classification.
+logger : Logger
+Defaults to `logging`.
+context : Context or list of Context
+Defaults to ``mx.cpu()``.
+work_load_list : list of number
+Default ``None``, indicating uniform workload.
+fixed_param_names: list of str
+Default ``None``, indicating no network parameters are fixed.
+state_names : list of str
+states are similar to data and label, but not provided by data
iterator.
+Instead they are initialized to 0 and can be set by `set_states()`.
+group2ctxs : dict of str to context or list of context,
+ or list of dict of str to context
+Default is `None`. Mapping the `ctx_group` attribute to the context
assignment.
+compression_params : dict
+Specifies type of gradient compression and additional arguments
depending
+on the type of compression being used. For example, 2bit compression
requires a threshold.
+Arguments would then be {'type':'2bit', 'threshold':0.5}
+See mxnet.KVStore.set_gradient_compression method for more details on
gradient compression.
+update_freq: int
+Specifies the number of times to update the full gradients to be used
in the SVRG optimization. For instance,
+update_freq = 2 will calculates the gradients over all data every two
epochs
+Examples
+
+>>> # An example of declaring and using SVRGModule.
+>>> mod = mod = SVRGModule(symbol=lro, data_names=['data'],
label_names=['lin_reg_label'], update_freq=2)
+>>> mod.fit(di, eval_metric='mse', optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),),
+>>> num_epoch=num_epoch, kvstore='local')
+"""
+
+def __init__(self, symbol, data_names=('data',),
label_names=('softmax_label',),
+ logger=logging, context=mx.cpu(), work_load_list=None,
+ fixed_param_names=None, state_names=None, group2ctxs=None,
+ compression_params=None, update_freq=None):
+super(SVRGModule, self).__init__(symbol, data_names=data_names,
label_names=label_names, logger=logger,
+ context=context,
work_load_list=work_load_list,
+ fixed_param_names=fixed_param_names,
state_names=state_names,
+ group2ctxs=group2ctxs,
compression_params=compression_params)
+
+# Type check update_frequency
+if isinstance(update_freq, int):
+self.update_freq = update_freq
+else:
+raise TypeError("update_freq must be an integer")
+
+self._mod_aux = mx.mod.Module(symbol, data_names, label_names, logger,
context, work_load_list,
+ fixed_param_names, state_names,
group2ctxs, compression_params)
+
+self._param_dict = [{} for ctx in self._context]
+
+def _reset_bind(self):
+"""Internal function to reset binded state."""
+super(SVRGModule, self)._reset_bind()
+self._mod_aux._reset_bind()
+
Review comment:
[GitHub] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213413094 ## File path: contrib/svrg_optimization_python/src/svrg_optimizer.py ## @@ -0,0 +1,131 @@ +# 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. +"""A `SVRGOptimizer` encapsulates two optimizers to accommodate SVRG optimization logic. +""" + + +import mxnet as mx + + +@mx.optimizer.register +class AssignmentOptimizer(mx.optimizer.Optimizer): +def update(self, index, weight, grad, state): +weight[:] = grad + +@mx.optimizer.register +class SVRGOptimizer(mx.optimizer.Optimizer): +"""SVRGOptimizer is a wrapper class for two optimizers: one for accumulating full gradients and the other Review comment: This is mainly useful with SGD right? Why allow any optimizer to be passed in here? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213418335
##
File path: contrib/svrg_optimization_python/test_svrg_train.py
##
@@ -0,0 +1,84 @@
+# 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 mxnet as mx
+import numpy as np
+from src.svrg_module import SVRGModule
+
+
+def test_svrg_intermediate_level_api(num_epoch):
+"""Test intermediate level svrgmodule API where the training process
+need to be explicitly defined. KVstore is not explicitly created.
+"""
+di, mod = create_network()
+mod.bind(data_shapes=di.provide_data, label_shapes=di.provide_label)
+mod.init_params(initializer=mx.init.Uniform(0.01), allow_missing=False,
force_init=False, allow_extra=False)
+kv = mx.kv.create("local")
+mod.init_optimizer(kvstore=kv, optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),))
+metrics = mx.metric.create("mse")
+for e in range(num_epoch):
+metrics.reset()
+if e % (mod.update_freq) == 0:
+mod.update_full_grads(di)
+di.reset()
+for batch in di:
+mod.forward_backward(data_batch=batch)
+mod.update()
+mod.update_metric(metrics, batch.label)
+print('Epoch[%d] Time cost=%.3f', e, metrics.get())
+
+
+def test_svrg_high_level_api(num_epoch):
+"""Test high level svrgmodule API. KVStore is explicitly created.
+"""
+di, mod = create_network()
+mod.fit(di, eval_metric='mse', optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),), num_epoch=num_epoch,
+kvstore='local')
+
+
+def create_network():
+"""Create a linear regression network for performing SVRG optimization.
+:return: an instance of mx.io.NDArrayIter
+:return: an instance of mx.mod.svrgmodule for performing SVRG optimization
+"""
+mx.random.seed(42)
+train_data = np.random.randint(1, 5, [1000, 2])
+weights = np.array([1.0, 2.0])
+train_label = train_data.dot(weights)
+
+di = mx.io.NDArrayIter(train_data, train_label, batch_size=32,
shuffle=True, label_name='lin_reg_label')
+X = mx.sym.Variable('data')
+Y = mx.symbol.Variable('lin_reg_label')
+fully_connected_layer = mx.sym.FullyConnected(data=X, name='fc1',
num_hidden=1)
+lro = mx.sym.LinearRegressionOutput(data=fully_connected_layer, label=Y,
name="lro")
+
+mod = SVRGModule(
+symbol=lro,
+data_names=['data'],
+label_names=['lin_reg_label'], update_freq=2
+)
+
+return di, mod
+
+# run as a script
+if __name__ == "__main__":
Review comment:
should it be? Please see other tests.
```
if __name__ == '__main__':
import nose
nose.runmodule()
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213417944
##
File path: contrib/svrg_optimization_python/test_svrg_train.py
##
@@ -0,0 +1,84 @@
+# 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 mxnet as mx
+import numpy as np
+from src.svrg_module import SVRGModule
+
+
+def test_svrg_intermediate_level_api(num_epoch):
+"""Test intermediate level svrgmodule API where the training process
+need to be explicitly defined. KVstore is not explicitly created.
+"""
+di, mod = create_network()
+mod.bind(data_shapes=di.provide_data, label_shapes=di.provide_label)
+mod.init_params(initializer=mx.init.Uniform(0.01), allow_missing=False,
force_init=False, allow_extra=False)
+kv = mx.kv.create("local")
+mod.init_optimizer(kvstore=kv, optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),))
+metrics = mx.metric.create("mse")
+for e in range(num_epoch):
+metrics.reset()
+if e % (mod.update_freq) == 0:
+mod.update_full_grads(di)
+di.reset()
+for batch in di:
+mod.forward_backward(data_batch=batch)
+mod.update()
+mod.update_metric(metrics, batch.label)
+print('Epoch[%d] Time cost=%.3f', e, metrics.get())
+
+
+def test_svrg_high_level_api(num_epoch):
+"""Test high level svrgmodule API. KVStore is explicitly created.
+"""
+di, mod = create_network()
+mod.fit(di, eval_metric='mse', optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),), num_epoch=num_epoch,
+kvstore='local')
+
+
+def create_network():
+"""Create a linear regression network for performing SVRG optimization.
+:return: an instance of mx.io.NDArrayIter
+:return: an instance of mx.mod.svrgmodule for performing SVRG optimization
+"""
+mx.random.seed(42)
Review comment:
1. Please do not use fixed seed. You can see other tests under tests/
2. Also, what are we asserting here?
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213406018
##
File path: contrib/svrg_optimization_python/src/svrg_module.py
##
@@ -0,0 +1,581 @@
+# 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.
+"""A `SVRGModule` implements the `Module` API by wrapping an auxiliary module
to perform
+SVRG optimization logic.
+"""
+
+import mxnet as mx
+import time
+import logging
+from svrg_optimizer import SVRGOptimizer
+from mxnet.module import Module
+
+
+class SVRGModule(Module):
+"""SVRGModule is a module that encapsulates two Modules to accommodate the
SVRG optimization technique.
+It is functionally the same as Module API, except it is implemented using
SVRG optimization logic.
+
+Parameters
+--
+symbol : Symbol
+data_names : list of str
+Defaults to `('data')` for a typical model used in image
classification.
+label_names : list of str
+Defaults to `('softmax_label')` for a typical model used in image
+classification.
+logger : Logger
+Defaults to `logging`.
+context : Context or list of Context
+Defaults to ``mx.cpu()``.
+work_load_list : list of number
+Default ``None``, indicating uniform workload.
+fixed_param_names: list of str
+Default ``None``, indicating no network parameters are fixed.
+state_names : list of str
+states are similar to data and label, but not provided by data
iterator.
+Instead they are initialized to 0 and can be set by `set_states()`.
+group2ctxs : dict of str to context or list of context,
+ or list of dict of str to context
+Default is `None`. Mapping the `ctx_group` attribute to the context
assignment.
+compression_params : dict
+Specifies type of gradient compression and additional arguments
depending
+on the type of compression being used. For example, 2bit compression
requires a threshold.
+Arguments would then be {'type':'2bit', 'threshold':0.5}
+See mxnet.KVStore.set_gradient_compression method for more details on
gradient compression.
+update_freq: int
+Specifies the number of times to update the full gradients to be used
in the SVRG optimization. For instance,
+update_freq = 2 will calculates the gradients over all data every two
epochs
+Examples
+
+>>> # An example of declaring and using SVRGModule.
+>>> mod = mod = SVRGModule(symbol=lro, data_names=['data'],
label_names=['lin_reg_label'], update_freq=2)
Review comment:
Please correct this example.
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213414138 ## File path: contrib/svrg_optimization_python/test_svrg_train.py ## @@ -0,0 +1,84 @@ +# Licensed to the Apache Software Foundation (ASF) under one Review comment: tests should go tests. May be like below? 1. tests/python/train/test_contrib_svrg.py : Full end to end train and testing. See test_mlp.py as an example. Please, note, it should be faster (< 2 min at max?). If it is longer than that, we need to move it to nightly test. 2. tests/python/unittest/test_contrib_svrgmodule.py / ..._svrgoptimizer.py etc.. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on a change in pull request #12376: [MXNET-854]
SVRG Optimization in Python Module API
URL: https://github.com/apache/incubator-mxnet/pull/12376#discussion_r213406770
##
File path: contrib/svrg_optimization_python/src/svrg_module.py
##
@@ -0,0 +1,581 @@
+# 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.
+"""A `SVRGModule` implements the `Module` API by wrapping an auxiliary module
to perform
+SVRG optimization logic.
+"""
+
+import mxnet as mx
+import time
+import logging
+from svrg_optimizer import SVRGOptimizer
+from mxnet.module import Module
+
+
+class SVRGModule(Module):
+"""SVRGModule is a module that encapsulates two Modules to accommodate the
SVRG optimization technique.
+It is functionally the same as Module API, except it is implemented using
SVRG optimization logic.
+
+Parameters
+--
+symbol : Symbol
+data_names : list of str
+Defaults to `('data')` for a typical model used in image
classification.
+label_names : list of str
+Defaults to `('softmax_label')` for a typical model used in image
+classification.
+logger : Logger
+Defaults to `logging`.
+context : Context or list of Context
+Defaults to ``mx.cpu()``.
+work_load_list : list of number
+Default ``None``, indicating uniform workload.
+fixed_param_names: list of str
+Default ``None``, indicating no network parameters are fixed.
+state_names : list of str
+states are similar to data and label, but not provided by data
iterator.
+Instead they are initialized to 0 and can be set by `set_states()`.
+group2ctxs : dict of str to context or list of context,
+ or list of dict of str to context
+Default is `None`. Mapping the `ctx_group` attribute to the context
assignment.
+compression_params : dict
+Specifies type of gradient compression and additional arguments
depending
+on the type of compression being used. For example, 2bit compression
requires a threshold.
+Arguments would then be {'type':'2bit', 'threshold':0.5}
+See mxnet.KVStore.set_gradient_compression method for more details on
gradient compression.
+update_freq: int
+Specifies the number of times to update the full gradients to be used
in the SVRG optimization. For instance,
+update_freq = 2 will calculates the gradients over all data every two
epochs
+Examples
+
+>>> # An example of declaring and using SVRGModule.
+>>> mod = mod = SVRGModule(symbol=lro, data_names=['data'],
label_names=['lin_reg_label'], update_freq=2)
+>>> mod.fit(di, eval_metric='mse', optimizer='sgd',
optimizer_params=(('learning_rate', 0.025),),
+>>> num_epoch=num_epoch, kvstore='local')
+"""
+
+def __init__(self, symbol, data_names=('data',),
label_names=('softmax_label',),
+ logger=logging, context=mx.cpu(), work_load_list=None,
+ fixed_param_names=None, state_names=None, group2ctxs=None,
+ compression_params=None, update_freq=None):
+super(SVRGModule, self).__init__(symbol, data_names=data_names,
label_names=label_names, logger=logger,
+ context=context,
work_load_list=work_load_list,
+ fixed_param_names=fixed_param_names,
state_names=state_names,
+ group2ctxs=group2ctxs,
compression_params=compression_params)
+
+# Type check update_frequency
+if isinstance(update_freq, int):
+self.update_freq = update_freq
+else:
+raise TypeError("update_freq must be an integer")
Review comment:
Can you make error more useful for user to understand the issue. Example:
(update_freq in SVRGModule must be an integer. Example: 2. Given update_freq =
', update_freq)
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact
[GitHub] sandeep-krishnamurthy commented on issue #12376: [MXNET-854] SVRG Optimization in Python Module API
sandeep-krishnamurthy commented on issue #12376: [MXNET-854] SVRG Optimization in Python Module API URL: https://github.com/apache/incubator-mxnet/pull/12376#issuecomment-416673829 Thanks @StephanieYuan - Welcome to MXNet community! @piiswrong @eric-haibin-lin @anirudhacharya @Roshrini @vandanavk - You will be interested to have a look at this. This is an automated message from the Apache Git Service. To respond to the message, please log on 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] apeforest commented on issue #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on issue #11952: [MXNET-707] Add unit test for mxnet to coreml converter URL: https://github.com/apache/incubator-mxnet/pull/11952#issuecomment-416673642 @Roshrini Addressed all your comments. Please kindly review again. Thanks! This is an automated message from the Apache Git Service. To respond to the message, please log on 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] apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on a change in pull request #11952: [MXNET-707] Add unit
test for mxnet to coreml converter
URL: https://github.com/apache/incubator-mxnet/pull/11952#discussion_r213404931
##
File path: tools/coreml/unittest/test_converter_no_pred.py
##
@@ -0,0 +1,970 @@
+# 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 unittest
+import mxnet as mx
+import numpy as np
+
+from converter._mxnet_converter import convert
+from collections import namedtuple
+from converter import utils
+
+def _mxnet_remove_batch(input_data):
+for blob in input_data:
+input_data[blob] = np.reshape(input_data[blob],
input_data[blob].shape[1:])
+return input_data
+
+
+def _get_mxnet_module(net, data_shapes, mode, label_names, input_names=None):
+""" Given a symbolic graph, input shape and the initialization mode,
+returns an MXNet module.
+"""
+mx.random.seed(1993)
+
+mod = utils.create_module(sym=net, data_shapes=data_shapes,
label_shapes=input_names, label_names=label_names)
+
+if mode == 'random':
+mod.init_params(
+initializer=mx.init.Uniform(scale=.1)
+)
+elif mode == 'zeros':
+mod.init_params(
+initializer=mx.init.Zero()
+)
+elif mode == 'ones':
+mod.init_params(
+initializer=mx.init.One()
+)
+else:
+Exception(KeyError("%s is not a valid initialization mode" % mode))
+
+return mod
+
+
+class SingleLayerTest(unittest.TestCase):
+"""
+Unit test class for testing where converter is able to convert individual
layers or not.
+In order to do so, it converts model and generates preds on both CoreML
and MXNet and check they are the same.
+"""
+def _test_mxnet_model(self, net, input_shape, mode, class_labels=None,
coreml_mode=None, label_names=None, delta=1e-3,
+ pre_processing_args=None, input_name='data'):
+""" Helper method that convert the CoreML model into CoreML and
compares the predictions over random data.
+
+Parameters
+--
+net: MXNet Symbol Graph
+The graph that we'll be converting into CoreML.
+
+input_shape: tuple of ints
+The shape of input data. Generally of the format (batch-size,
channels, height, width)
+
+mode: (random|zeros|ones)
+The mode to use in order to set the parameters (weights and
biases).
+
+label_names: list of strings
+The names of the output labels. Default: None
+
+delta: float
+The maximum difference b/w predictions of MXNet and CoreML that is
tolerable.
+
+input_name: str
+The name of the input variable to the symbolic graph.
+"""
+
+data_shapes=[(input_name, input_shape)]
+
+mod = _get_mxnet_module(net, data_shapes, mode, label_names)
+
+# Generate some dummy data
+input_data = {input_name: np.random.uniform(-10., 10., input_shape)}
+Batch = namedtuple('Batch', ['data'])
+mod.forward(Batch([mx.nd.array(input_data[input_name])]))
+mxnet_preds = mod.get_outputs()[0].asnumpy().flatten()
+
+# Get predictions from coreml
+coreml_model = convert(
+model=mod,
+class_labels=class_labels,
+mode=coreml_mode,
+input_shape={input_name: input_shape},
+preprocessor_args=pre_processing_args
+)
+
+def test_tiny_inner_product_zero_input(self):
+np.random.seed(1988)
+input_shape = (1, 10)
+net = mx.sym.Variable('data')
+net = mx.sym.FullyConnected(data=net, name='fc1', num_hidden=5)
+self._test_mxnet_model(net, input_shape=input_shape, mode='zeros')
+
+def test_really_tiny_inner_product_ones_input(self):
+np.random.seed(1988)
+input_shape = (1, 1)
+net = mx.sym.Variable('data')
+net = mx.sym.FullyConnected(data=net, name='fc1', num_hidden=1)
+self._test_mxnet_model(net, input_shape=input_shape, mode='ones')
+
+def test_really_tiny_2_inner_product_ones_input(self):
+np.random.seed(1988)
+input_shape = (1, 1)
+net =
[GitHub] apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on a change in pull request #11952: [MXNET-707] Add unit
test for mxnet to coreml converter
URL: https://github.com/apache/incubator-mxnet/pull/11952#discussion_r213404849
##
File path: tools/coreml/unittest/test_converter_no_pred.py
##
@@ -0,0 +1,970 @@
+# 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 unittest
+import mxnet as mx
+import numpy as np
+
+from converter._mxnet_converter import convert
+from collections import namedtuple
+from converter import utils
+
+def _mxnet_remove_batch(input_data):
+for blob in input_data:
+input_data[blob] = np.reshape(input_data[blob],
input_data[blob].shape[1:])
+return input_data
+
+
+def _get_mxnet_module(net, data_shapes, mode, label_names, input_names=None):
+""" Given a symbolic graph, input shape and the initialization mode,
+returns an MXNet module.
+"""
+mx.random.seed(1993)
+
+mod = utils.create_module(sym=net, data_shapes=data_shapes,
label_shapes=input_names, label_names=label_names)
+
+if mode == 'random':
+mod.init_params(
+initializer=mx.init.Uniform(scale=.1)
+)
+elif mode == 'zeros':
+mod.init_params(
+initializer=mx.init.Zero()
+)
+elif mode == 'ones':
+mod.init_params(
+initializer=mx.init.One()
+)
+else:
+Exception(KeyError("%s is not a valid initialization mode" % mode))
+
+return mod
+
+
+class SingleLayerTest(unittest.TestCase):
+"""
+Unit test class for testing where converter is able to convert individual
layers or not.
+In order to do so, it converts model and generates preds on both CoreML
and MXNet and check they are the same.
+"""
+def _test_mxnet_model(self, net, input_shape, mode, class_labels=None,
coreml_mode=None, label_names=None, delta=1e-3,
+ pre_processing_args=None, input_name='data'):
+""" Helper method that convert the CoreML model into CoreML and
compares the predictions over random data.
+
+Parameters
+--
+net: MXNet Symbol Graph
+The graph that we'll be converting into CoreML.
+
+input_shape: tuple of ints
+The shape of input data. Generally of the format (batch-size,
channels, height, width)
+
+mode: (random|zeros|ones)
+The mode to use in order to set the parameters (weights and
biases).
+
+label_names: list of strings
+The names of the output labels. Default: None
+
+delta: float
+The maximum difference b/w predictions of MXNet and CoreML that is
tolerable.
+
+input_name: str
+The name of the input variable to the symbolic graph.
+"""
+
+data_shapes=[(input_name, input_shape)]
+
+mod = _get_mxnet_module(net, data_shapes, mode, label_names)
+
+# Generate some dummy data
+input_data = {input_name: np.random.uniform(-10., 10., input_shape)}
+Batch = namedtuple('Batch', ['data'])
+mod.forward(Batch([mx.nd.array(input_data[input_name])]))
+mxnet_preds = mod.get_outputs()[0].asnumpy().flatten()
+
+# Get predictions from coreml
+coreml_model = convert(
+model=mod,
+class_labels=class_labels,
+mode=coreml_mode,
+input_shape={input_name: input_shape},
+preprocessor_args=pre_processing_args
+)
+
+def test_tiny_inner_product_zero_input(self):
+np.random.seed(1988)
+input_shape = (1, 10)
+net = mx.sym.Variable('data')
+net = mx.sym.FullyConnected(data=net, name='fc1', num_hidden=5)
+self._test_mxnet_model(net, input_shape=input_shape, mode='zeros')
+
+def test_really_tiny_inner_product_ones_input(self):
Review comment:
Added for all
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] vishaalkapoor edited a comment on issue #12363: distributed training notebook tests
vishaalkapoor edited a comment on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416670681 @sandeep-krishnamurthy It looks like the tutorial/notebook executor will need to be wrapped by a launcher script that executes notebooks in parallel on several nodes. I modified the existing tutorial/notebook executor so that it could be used for more than tutorials (arbitrary notebooks) on a single-host. There is no scaffolding for multi-host. https://github.com/apache/incubator-mxnet/blob/ae5d60fa830090f4882a433d9b88c53c26c42b4f/tests/utils/notebook_test/__init__.py#L39 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] vishaalkapoor commented on issue #12363: distributed training notebook tests
vishaalkapoor commented on issue #12363: distributed training notebook tests URL: https://github.com/apache/incubator-mxnet/issues/12363#issuecomment-416670681 @sandeep-krishnamurthy It looks like the tutorial/notebook executor will need to be wrapped by a launcher script that executes notebooks in parallel on several nodes. I modified the existing tutorial/notebook executor so that it could be used for more than tutorials on a single-host. There is no scaffolding for multi-host. https://github.com/apache/incubator-mxnet/blob/ae5d60fa830090f4882a433d9b88c53c26c42b4f/tests/utils/notebook_test/__init__.py#L39 This is an automated message from the Apache Git Service. To respond to the message, please log on 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] sandeep-krishnamurthy commented on issue #11890: Error During Make - undefined reference
sandeep-krishnamurthy commented on issue #11890: Error During Make - undefined reference URL: https://github.com/apache/incubator-mxnet/issues/11890#issuecomment-416670461 @liangxi627 - can you please give more details? Is it due to version mis-match of OpenCV? This is an automated message from the Apache Git Service. To respond to the message, please log on 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] apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on a change in pull request #11952: [MXNET-707] Add unit
test for mxnet to coreml converter
URL: https://github.com/apache/incubator-mxnet/pull/11952#discussion_r213400743
##
File path: tools/coreml/unittest/test_converter_no_pred.py
##
@@ -0,0 +1,970 @@
+# 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 unittest
+import mxnet as mx
+import numpy as np
+
+from converter._mxnet_converter import convert
+from collections import namedtuple
+from converter import utils
+
+def _mxnet_remove_batch(input_data):
+for blob in input_data:
+input_data[blob] = np.reshape(input_data[blob],
input_data[blob].shape[1:])
+return input_data
+
+
+def _get_mxnet_module(net, data_shapes, mode, label_names, input_names=None):
+""" Given a symbolic graph, input shape and the initialization mode,
+returns an MXNet module.
+"""
+mx.random.seed(1993)
+
+mod = utils.create_module(sym=net, data_shapes=data_shapes,
label_shapes=input_names, label_names=label_names)
+
+if mode == 'random':
+mod.init_params(
+initializer=mx.init.Uniform(scale=.1)
+)
+elif mode == 'zeros':
+mod.init_params(
+initializer=mx.init.Zero()
+)
+elif mode == 'ones':
+mod.init_params(
+initializer=mx.init.One()
+)
+else:
+Exception(KeyError("%s is not a valid initialization mode" % mode))
+
+return mod
+
+
+class SingleLayerTest(unittest.TestCase):
+"""
+Unit test class for testing where converter is able to convert individual
layers or not.
+In order to do so, it converts model and generates preds on both CoreML
and MXNet and check they are the same.
+"""
+def _test_mxnet_model(self, net, input_shape, mode, class_labels=None,
coreml_mode=None, label_names=None, delta=1e-3,
+ pre_processing_args=None, input_name='data'):
+""" Helper method that convert the CoreML model into CoreML and
compares the predictions over random data.
+
+Parameters
+--
+net: MXNet Symbol Graph
+The graph that we'll be converting into CoreML.
+
+input_shape: tuple of ints
+The shape of input data. Generally of the format (batch-size,
channels, height, width)
+
+mode: (random|zeros|ones)
+The mode to use in order to set the parameters (weights and
biases).
+
+label_names: list of strings
+The names of the output labels. Default: None
+
+delta: float
+The maximum difference b/w predictions of MXNet and CoreML that is
tolerable.
+
+input_name: str
+The name of the input variable to the symbolic graph.
+"""
+
+data_shapes=[(input_name, input_shape)]
+
+mod = _get_mxnet_module(net, data_shapes, mode, label_names)
+
+# Generate some dummy data
+input_data = {input_name: np.random.uniform(-10., 10., input_shape)}
+Batch = namedtuple('Batch', ['data'])
+mod.forward(Batch([mx.nd.array(input_data[input_name])]))
+mxnet_preds = mod.get_outputs()[0].asnumpy().flatten()
+
+# Get predictions from coreml
+coreml_model = convert(
+model=mod,
+class_labels=class_labels,
+mode=coreml_mode,
+input_shape={input_name: input_shape},
+preprocessor_args=pre_processing_args
+)
+
+def test_tiny_inner_product_zero_input(self):
+np.random.seed(1988)
Review comment:
removed
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on a change in pull request #11952: [MXNET-707] Add unit
test for mxnet to coreml converter
URL: https://github.com/apache/incubator-mxnet/pull/11952#discussion_r213400631
##
File path: tools/coreml/unittest/test_converter_no_pred.py
##
@@ -0,0 +1,970 @@
+# 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 unittest
+import mxnet as mx
+import numpy as np
+
+from converter._mxnet_converter import convert
+from collections import namedtuple
+from converter import utils
+
+def _mxnet_remove_batch(input_data):
+for blob in input_data:
+input_data[blob] = np.reshape(input_data[blob],
input_data[blob].shape[1:])
+return input_data
+
+
+def _get_mxnet_module(net, data_shapes, mode, label_names, input_names=None):
+""" Given a symbolic graph, input shape and the initialization mode,
+returns an MXNet module.
+"""
+mx.random.seed(1993)
+
+mod = utils.create_module(sym=net, data_shapes=data_shapes,
label_shapes=input_names, label_names=label_names)
+
+if mode == 'random':
+mod.init_params(
+initializer=mx.init.Uniform(scale=.1)
+)
+elif mode == 'zeros':
+mod.init_params(
+initializer=mx.init.Zero()
+)
+elif mode == 'ones':
+mod.init_params(
+initializer=mx.init.One()
+)
+else:
+Exception(KeyError("%s is not a valid initialization mode" % mode))
+
+return mod
+
+
+class SingleLayerTest(unittest.TestCase):
+"""
+Unit test class for testing where converter is able to convert individual
layers or not.
Review comment:
Thanks. Corrected.
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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] apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter URL: https://github.com/apache/incubator-mxnet/pull/11952#discussion_r213400352 ## File path: tools/coreml/unittest/test_converter_no_pred.py ## @@ -0,0 +1,970 @@ +# 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 unittest +import mxnet as mx +import numpy as np + +from converter._mxnet_converter import convert +from collections import namedtuple +from converter import utils + +def _mxnet_remove_batch(input_data): Review comment: removed This is an automated message from the Apache Git Service. To respond to the message, please log on 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] apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter
apeforest commented on a change in pull request #11952: [MXNET-707] Add unit test for mxnet to coreml converter URL: https://github.com/apache/incubator-mxnet/pull/11952#discussion_r213400138 ## File path: tools/coreml/unittest/test_converter_no_pred.py ## @@ -0,0 +1,970 @@ +# 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. + Review comment: What docstring? This seems to be the standard header for all other python unittest. Please clarify. Thanks This is an automated message from the Apache Git Service. To respond to the message, please log on 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
