Thanks @andrewfayres for the detailed explanation. Next time please add the results also in the PR. What makes such PR hard is that we are basing off of manual tests which are not reproducible unless another person does the exact same manual test to verify. The concern I have in such cases is the possibility of introducing regressions somewhere else.
[ Full content available at: https://github.com/apache/incubator-mxnet/pull/11885 ] This message was relayed via gitbox.apache.org for [email protected]
