DickJC123 commented on issue #14954: broadcast_to() behavior change after new 0D and uninitialized shape support added? URL: https://github.com/apache/incubator-mxnet/issues/14954#issuecomment-492803312 I've been running in the top-level sockeye directory `python setup.py test`. I saw 22 unit tests fail. After changing line 392 in the above-posted snippet to: ``` if (oshape[i] != -1 && oshape[i] != 0) { ``` I'm down to 5 tests failing. I assume you're asking about these 5? They are: ``` test/unit/test_data_io.py::test_parallel_data_set FAILED [ 39%] test/unit/test_data_io.py::test_sharded_parallel_sample_iter FAILED [ 43%] test/integration/test_seq_copy_int.py::test_seq_copy ... FAILED [ 97%] test/integration/test_seq_copy_int.py::test_seq_copy ... FAILED [ 97%] test/integration/test_seq_copy_int.py::test_seq_copy ...FAILED [ 97%] ``` In each case the error is a device_type == 0 (neither cpu nor gpu) on a read-in NDArray. It's reading in a file that has neither of the 2 magic numbers (so 'legacy' loading) and getting confused (I think) on the reading in of a shape.
---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
