Importing model generated by AutoML Tables to BigQuery ML raises error "Op type not registered 'DecodeProtoSparseV2'"

Question

I am trying to import a TensorFlow model generated by AutoML Tables to BigQuery ML.

Unfortunately it gives me the following error

Op type not registered 'DecodeProtoSparseV2'

Does anyone know how to solve this issue?

Answer 1

Currently, the model exported from AutoML Tables can only run with the provided model server container. See this question: Loading model generated by AutoML with Tensorflow raises "Op type not registered 'ParseExampleV2" for more detail.

Update: BigQuery ML now supports AutoML Tables integration (in beta)

Answer 2

It seems to be an incompatibility with the exported model from AutoML Tables and Bigquery ML, I have replicated this scenario and I have got the same error

Error when loading TensorFlow SavedModel with tag set 'serve': Op type not registered 'DecodeProtoSparseV2' in binary running on sandbox2. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (eg) tf.contrib.resampler should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed.

Based on this, the issue seems to be related of how AutoML Tables is exporting the model. Another point to consider is that Bigquery ML has the next limitations

On the other hand, I would be useful to know, why do you want to use the AutoML Tables model over Bigquery? It's not clear the use case since AutoML has some advantages that could avoid using Bigquery ML

Answer 3

Try to install and import struct2tensor

https://pypi.org/project/struct2tensor/