huyangc opened a new issue #9911: The different between rescale_grad on optimizer and normalization in loss layer. URL: https://github.com/apache/incubator-mxnet/issues/9911 When I normalization the gradient of the SoftmaxOutput using ``normalization=valid``, and setting the ``rescale_grad=1/batchsize``, it seems like that the gradient will be rescale twice? First is using the number of valid label and second using the batchsize?
---------------------------------------------------------------- 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
