Issue deploying AutoML model generated in BigQuery to AI Platform

0
votes

I am having trouble deploying an AutoML model created using BigQuery to AI Platform for online prediction.

I have created an AutoML model in BigQuery using the standard procedure via SQL:

CREATE OR REPLACE MODEL `model_name`
OPTIONS
  (model_type='automl_regressor', budget_hours=2.0,
    ... ) AS
SELECT ...)

This works fine and I am able to get predicted results successfully. I now wanted to deploy it for online prediction. To do so I exported the model to a GCS bucket via the Export Model function in the BigQuery Cloud Console. This gives me a directory in the bucket with the following contents:

assets/
saved_model.pb
variables/

I then went to the AI Platform console and created a Model and then proceeded to create a Version for that model with the following pre-built container settings:

  • Python Version: 3.7
  • Framework: TensorFlow
  • Framework Version: 2.3.1
  • ML runtime version: 2.3

I have set the Cloud Storage path to the bucket with the directory containing the contents that I listed above and proceeded to create the Version for my Model. Upon doing so I get this error after some time:

Create Version failed. Bad model detected with error: "Failed to load model: Loading servable: {name: default version: 1} failed: Not found: Op type not registered 'DecodeProtoSparseV2' in binary running on localhost. 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 (e.g.) `tf.contrib.resampler` should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed.\n\n (Error code: 0)"

Kind of stumped here as I thought this was the way I could utilize the ML model generated from BigQuery. Is there anything wrong with the steps here? Is it even possible to deploy such a model for online prediction currently? If not, is there a way I can convert the model so that it can be deployed? Any help would be appreciated!

2
tensorflowgoogle-cloud-platformgoogle-bigquerygoogle-cloud-automlgoogle-ai-platform

2 Answers

1
votes

As I understand you have trained an automl_regressor model in BigQuery ML and you aim to use it for online prediction with AI Platform. However, according to the documentation, it is not possible to use automl_regressor for online predictions.

  • Exported models for model types AUTOML_REGRESSOR and AUTOML_CLASSIFIER do not support AI Platform deployment for online prediction.

Depending on your needs, you can choose another model from BigQuery ML list here, which can be deployed for online prediction.

Furthermore, you can also train your model directly with AutoML and deploy and use it for online predictions. Although, please notice that AutoML models are used for specific prediction requirements such as described here. Also, you need to follow the documentation for training the model.

1
votes

You can also use the AutoML container/Docker to deploy an AutoML model trained in BigQuery using the documentation here. This is similar to how you would export and deploy a model trained directly using AutoML Tables.