You are viewing a plain text version of this content. The canonical link for it is here.

Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2020/07/10 19:46:26 UTC

[GitHub] [incubator-mxnet] DickJC123 commented on pull request #18688: Fix the flaky test 'test_npx_batch_norm'

DickJC123 commented on pull request #18688:
URL: https://github.com/apache/incubator-mxnet/pull/18688#issuecomment-656859038


   I ran into this issue developing a PR I have yet to submit.  The discrepancy in model outputs is caused by the fact that when cudnn calculates the running variance, it uses the 'sample variance', while this test is comparing in all cases to the 'population variance'.  The difference is that the sample variance uses a factor of N-1 in the denominator, while the population variance uses N (where N is the number of elements in the sample).
   
   My upcoming PR will include a fix for this, and after it's merged, if you want you could revert this commit that changed the problem sizes, since that is not the real issue here.


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org