Can I use exported AutoML Tables model in Python?

Question

Can I export a model from AutoML tables and use it locally with tensorflow?

I've discovered that you can export some of the Vision models in multiple formats but for Tables I can only export a tf edges model to Cloud Storage. So can I run it locally somehow to perform predictions? Or is there another way to export/convert a model in a format that can be used locally?

Answer 1

It is possible for AutoML tables to export a tf saved model (.pb format). You can do this by sending a request using curl. See Exporting Models in AutoML Tables on detailed instructions on how to use models.export.

NOTE: Exported model used for testing is from AutoML Tables quickstart.

JSON request (Use tf_saved_model to export a tensorflow model in SavedModel format):

{
  "outputConfig": {
    "modelFormat": "tf_saved_model",
    "gcsDestination": {
      "outputUriPrefix": "your-gcs-destination"
    }
  }
}

Curl command (assuming that the model is in us-central1):

curl -X POST \
-H "Authorization: Bearer "$(gcloud auth application-default print-access-token) \
-H "Content-Type: application/json; charset=utf-8" \
-d @request.json \
https://automl.googleapis.com/v1beta1/projects/your-project-id/locations/us-central1/models/your-model-id:export

Exported model: