When installing feast sdk in an environment with Tensorflow Model Analysis, TFMA will stop working due to the fact that feast overwrites the files on the tensorflow_metadata python package.
Feast overwrites tf_metadata files on pip install.
tensorflow_metadata/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
tensorflow_metadata/__pycache__/__init__.cpython-37.pyc,,
tensorflow_metadata/proto/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
tensorflow_metadata/proto/__pycache__/__init__.cpython-37.pyc,,
tensorflow_metadata/proto/v0/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
tensorflow_metadata/proto/v0/__pycache__/__init__.cpython-37.pyc,,
tensorflow_metadata/proto/v0/__pycache__/path_pb2.cpython-37.pyc,,
tensorflow_metadata/proto/v0/__pycache__/schema_pb2.cpython-37.pyc,,
tensorflow_metadata/proto/v0/__pycache__/statistics_pb2.cpython-37.pyc,,
tensorflow_metadata/proto/v0/path_pb2.py,sha256=8f1IIPnwSr0TUeaBbLrrA0SOvVtWZmJuekClUK0fohg,2341
tensorflow_metadata/proto/v0/path_pb2.pyi,sha256=x7plHZ0kvkv5qT8mZTJk8yqlQHdNdyAQM52GrHDdsU4,1341
tensorflow_metadata/proto/v0/schema_pb2.py,sha256=mnnrq9bVYIP5Uz9ztYoIDknhJAiJN2o8NuAonroLUUk,107886
tensorflow_metadata/proto/v0/schema_pb2.pyi,sha256=We0EW9WdmjfXRkDgx_l_UHW3WyFnSu03tSxRJY2R1xw,41992
tensorflow_metadata/proto/v0/statistics_pb2.py,sha256=TtNbajCCORqqIrl-KIodj4gwjjwA_w4NOvcZaSGGfcA,87569
tensorflow_metadata/proto/v0/statistics_pb2.pyi,sha256=fQk1JPNKKwoR9PVnlKrWOQo7AnpxDQpTiiUVmot_gIk,31328
Ideally this should be instead a python dependency for feast so that tf_metadata can be managed as a python dependency.
Add tensorflow-metadata as a dependency of the python sdk instead. Allowing users to control the tf_metadata version installed on their system.
Expected Behavior
When installing feast sdk in an environment with Tensorflow Model Analysis, TFMA will stop working due to the fact that feast overwrites the files on the tensorflow_metadata python package.
This generates issues like this: https://github.com/BACtaki/tfx-addons/runs/4196630120?check_suite_focus=true
Current Behavior
Feast overwrites tf_metadata files on pip install.
Steps to reproduce
Just open feast wheel and inspect the RECORD file you get:
Ideally this should be instead a python dependency for feast so that tf_metadata can be managed as a python dependency.
Specifications
Possible Solution
Add tensorflow-metadata as a dependency of the python sdk instead. Allowing users to control the tf_metadata version installed on their system.