[GitHub] eric-haibin-lin opened a new issue #7317: check_numeric_grad with FullyConnected fails

[gi...@git.apache.org]

eric-haibin-lin opened a new issue #7317: check_numeric_grad with FullyConnected fails
URL: https://github.com/apache/incubator-mxnet/issues/7317
 
 
   For bugs or installation issues, please provide the following information.
   The more information you provide, the more likely people will be able to help you.
   
   ## Environment info
   Operating System: DL AMI
   
   Compiler:
   
   Package used (Python/R/Scala/Julia):
   
   MXNet version:
   
   Or if installed from source: 
   
   MXNet commit hash (`git rev-parse HEAD`): 861e929cec8fa8fbab06884d9605debb74cd7217
   
   If you are using python package, please provide
   
   Python version and distribution:
   
   If you are using R package, please provide
   
   R `sessionInfo()`:
   
   ## Error Message:
   Please paste the full error message, including stack trace.
   ```
   /home/ubuntu/upstream-cpu/python/mxnet/test_utils.py:412: UserWarning: Calling forward the second time after forward(is_train=True) without calling backward first. Is this intended$
     executor.forward(is_train=use_forward_train)
   /home/ubuntu/upstream-cpu/python/mxnet/test_utils.py:404: UserWarning: Calling forward the second time after forward(is_train=True) without calling backward first. Is this intended?
     executor.forward(is_train=use_forward_train)
   Traceback (most recent call last):
     File "./test.py", line 13, in <module>
       check_fully_connected((32, 100), 10)
     File "./test.py", line 11, in check_fully_connected
       check_numeric_gradient(test, location)
     File "/home/ubuntu/upstream-cpu/python/mxnet/test_utils.py", line 524, in check_numeric_gradient
       ("NUMERICAL_%s"%name, "BACKWARD_%s"%name))
     File "/home/ubuntu/upstream-cpu/python/mxnet/test_utils.py", line 154, in assert_almost_equal
       raise AssertionError(msg)
   AssertionError:
   Items are not equal:
   Error 294.098633 exceeds tolerance rtol=0.010000, atol=0.000000.  Location of maximum error:(19, 54), a=0.001162, b=0.000295
    NUMERICAL_x: array([[ 3.73460722,  0.2527386 , -1.75657499, ..., -1.24495852,
            0.74711072, -5.7607851 ],
          [ 2.94259191, -0.7109642 , -1.08036399, ..., -0.87466836,...
    BACKWARD_x: array([[ 3.7337687 ,  0.25231028, -1.75907421, ..., -1.24670959,
            0.74810255, -5.76152706],
          [ 2.94111443, -0.7087307 , -1.0836978 , ..., -0.87180936,...
   ```
   ## Minimum reproducible example
   if you are using your own code, please provide a short script that reproduces the error.
   
   ```
   import mxnet as mx
   from mxnet.test_utils import *
   def check_fully_connected(shape, num_hidden):
       x = mx.symbol.Variable('x')
       w = mx.symbol.Variable('w')
       test = mx.sym.FullyConnected(data=x, weight=w, num_hidden=num_hidden, no_bias=True)
       x_nd = mx.nd.array(random_arrays(shape))
       w_nd = mx.nd.array(random_arrays((num_hidden, shape[1])))
       location = {'x': x_nd, 'w': w_nd}
       out_dns = mx.nd.dot(x_nd, w_nd.T).asnumpy()
       check_numeric_gradient(test, location)
   
   check_fully_connected((32, 100), 10)
   ```
   
   ## Steps to reproduce
   or if you are running standard examples, please provide the commands you have run that lead to the error.
   
   1. run the above code
   2.
   3.
   
   ## What have you tried to solve it?
   
   1. 
   2.
   3.
   
 
----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services