metadata
metadata copied to clipboard
Missing files in 0.12.0 release (ImportError: cannot import name schema_pb2)
I believe there has been some issue with the packaging because I am getting this error when running a pipeline with tf.Transform and getting the following error that after some investigation led me to tensorflow-metadata
File "/usr/local/lib/python2.7/dist-packages/tensorflow_transform/init.py", line 21, in
from tensorflow_transform.mappers import * File "/usr/local/lib/python2.7/dist-packages/tensorflow_transform/mappers.py", line 62, in from tensorflow_transform import schema_inference File "/usr/local/lib/python2.7/dist-packages/tensorflow_transform/schema_inference.py", line 30, in from tensorflow_transform.tf_metadata import dataset_schema File "/usr/local/lib/python2.7/dist-packages/tensorflow_transform/tf_metadata/dataset_schema.py", line 23, in from tensorflow_transform.tf_metadata import schema_utils File "/usr/local/lib/python2.7/dist-packages/tensorflow_transform/tf_metadata/schema_utils.py", line 28, in from tensorflow_metadata.proto.v0 import schema_pb2 ImportError: cannot import name schema_pb2
I checked the files at https://pypi.org/project/tensorflow-metadata/#files and they are indeed missing protos that should be there as per the aforementioned ImportError
There seems to be a packaging issue in the latest version. Can you pin the tensorflow-metadata version to be 0.9.0 for now? We will do a minor release to fix this.
Yes, pinning to the previous version solves the problem for now until the current package is pulled and/or fixed.
A new version of tensorflow-metadata (0.12.1) with the fix has been released.
Hey @paulgc, appreciate the quick fix.
It looks like 0.12.1
release missed a few things though. There is no release tag and also the __version__
in setup.py
is still 0.12.0
. https://github.com/tensorflow/metadata/blob/42939154a6ef763c9a41fed48abea758df7c0d6a/setup.py#L20
@mzinkevi @solma
This issue seems still reproducible on x86_64 and arm64 both platforms while testing with the "pytest" command. Please share if you have any plan to fix this in some upcoming releases.