You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
ProcessPool uses pyarrow.serialize. However, it would map all integer fields to int32 type. Tensorflow requires exact types to be returned by a pyfunc, hence it will fail.
We did not see this issue since we are not covering the combination of ProcessPool and tf_tensors in test_tf_utils.py.
ProcessPool is going to be replaced when we fully adapt Reader v2, so probably do not have to fix the issue. Opening this issue for awareness.
The text was updated successfully, but these errors were encountered:
selitvin
changed the title
tf_tensors would fail when reading a dataset with non int32 fields and ProcessPool implementation is used
tf_tensors would fail when reading a dataset with non int32 fields and ProcessPool(pyarrow_serialize=True) implementation is used
Oct 22, 2018
ProcessPool uses
pyarrow.serialize
. However, it would map all integer fields to int32 type. Tensorflow requires exact types to be returned by a pyfunc, hence it will fail.We did not see this issue since we are not covering the combination of ProcessPool and tf_tensors in
test_tf_utils.py
.ProcessPool is going to be replaced when we fully adapt Reader v2, so probably do not have to fix the issue. Opening this issue for awareness.
The text was updated successfully, but these errors were encountered: