Zha0q1 commented on issue #15069: [WIP] Profiler RFC: Introducing new APIs
URL:
https://github.com/apache/incubator-mxnet/issues/15069#issuecomment-497415990
> do you feel that aggregating raw profiler dumps is worth adding a new C
API? to me, this aggregation doesn't seem like a core
Zha0q1 commented on issue #15069: [WIP] Profiler RFC: Introducing new APIs
URL:
https://github.com/apache/incubator-mxnet/issues/15069#issuecomment-497072274
I think there are multiple points:
1. People are already using dumps() and we want to present the information
in a way (json)
Zha0q1 commented on issue #15069: [WIP] Profiler RFC: Introducing new APIs
URL:
https://github.com/apache/incubator-mxnet/issues/15069#issuecomment-497058193
> > So instead of adding a utility that parse that table into a JSON string,
we can return a JSON string from the backend directly.
Zha0q1 commented on issue #15069: [WIP] Profiler RFC: Introducing new APIs
URL:
https://github.com/apache/incubator-mxnet/issues/15069#issuecomment-496808663
> sorting for the tabular output would be great to have! There is already an
issue open for this: #14100
Thanks for the
Zha0q1 commented on issue #15069: [WIP] Profiler RFC: Introducing new APIs
URL:
https://github.com/apache/incubator-mxnet/issues/15069#issuecomment-496783841
> > Why add a new profiler API get_summary() in the back-end rather than a
python parser utility that returns in JSON?
> > This
Zha0q1 commented on issue #15069: [WIP] Profiler RFC: Introducing new APIs
URL:
https://github.com/apache/incubator-mxnet/issues/15069#issuecomment-496782815
> I understand the value of getting the dump output in a different format
compared to tabular form. This can also be useful for