marcoabreu commented on issue #11544: [MXNET-620]Fix flaky test batchnorm 
training
URL: https://github.com/apache/incubator-mxnet/pull/11544#issuecomment-403205212
 
 
   I think we should leave the tests as they are and there can be additional 
specific tests for the mkldnn backend. 
   
   Our users should not have to care what backend is being selected and the 
behaviour should always be identical. This means that however I call a certain 
operator, row_sparse in this case, I would expect that I get a proper result in 
**any** case. It's great that the issue is already tracked, but I'm heavily 
opposed to removing that operator from our test suite. All operator tests have 
to be implementation independent and should not differentiate or exclude 
certain cases. From my point of view, this flaky test is only resolved when the 
mkldnn implementation has been fixed and when this test passes all the time. 
Removing it is not an option.

----------------------------------------------------------------
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:
[email protected]


With regards,
Apache Git Services

Reply via email to