[GitHub] [incubator-mxnet] lebeg commented on issue #15374: NightlyTestsForBinaries tutorials test broken

[GitBox <gi...@apache.org>]

lebeg commented on issue #15374: NightlyTestsForBinaries tutorials test broken
URL: https://github.com/apache/incubator-mxnet/issues/15374#issuecomment-505971155
 
 
   The test was broken by this PR [#15170: [MXNET-1413] Adding Large Tensor support for sort operators](https://github.com/apache/incubator-mxnet/pull/15170).
   The reason is that the default type of NDArray for `ArgSort` operator [changed](https://github.com/apache/incubator-mxnet/pull/15170/files#diff-067178695b7fe40b05ad18fa1a25c69dR91
   ) from `kFloat32` to `kInt64` or `kInt32`.
   ```cpp
   #if USE_INT64_TENSOR_SIZE == 1
       .set_default(mshadow::kInt64)
   #else
       .set_default(mshadow::kInt32)
   #endif
   ```
   This has caused a problem in [gluon-cv](https://github.com/dmlc/gluon-cv) in [`SSDMultiBoxLoss`](https://github.com/dmlc/gluon-cv/blob/c2296f87484b204db4c563987db819c768ae51c6/gluoncv/loss.py#L99) in particular at [this line](https://github.com/dmlc/gluon-cv/blob/c2296f87484b204db4c563987db819c768ae51c6/gluoncv/loss.py#L157)
   
   ```python
   rank = (cls_loss * (pos - 1)).argsort(axis=1).argsort(axis=1)
   ```
   

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


With regards,
Apache Git Services