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 2018/11/01 04:02:00 UTC

[GitHub] leezu commented on issue #13059: Refactor L2_normalization

leezu commented on issue #13059: Refactor L2_normalization
URL: https://github.com/apache/incubator-mxnet/pull/13059#issuecomment-434923748
 
 
   Thanks for your PR! Your current implementation will tend to underflow for small values and overflow for large values. How about using a stable algorithm for normalizing? You can take a look at https://github.com/apache/incubator-mxnet/blob/2899715921612ef4dd147004292b5b5d0f83320b/src/operator/mshadow_op.h#L892 or the BLAS nrm2 implementation. There is an open PR that changes L2_normalization Op to make use of the stable implementation, but it is currently abandoned https://github.com/apache/incubator-mxnet/pull/12440

----------------------------------------------------------------
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