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 2020/06/15 10:04:28 UTC

[GitHub] [incubator-mxnet] wkcn commented on issue #18563: The gradient of a tensor whose elements are all the same is wrong when the tensor pass to nd.max

wkcn commented on issue #18563:
URL: https://github.com/apache/incubator-mxnet/issues/18563#issuecomment-644033481


   Thank you for the report!
   I think it is a bug.
   
   MXNet use the function [`ReduceAxesBackwardUseInOut<cpu, mshadow_op::eq>`](https://github.com/apache/incubator-mxnet/blob/master/src/operator/tensor/broadcast_reduce_minmax_value.cc#L43) to compute the gradient of `min` and `max`. It means that the input gradient of the location, whose output is equal to the minimal value or the maximum value, is 1. It is wrong.


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