Loading model generated by AutoML with Tensorflow raises "Op type not registered 'ParseExampleV2"

Question

So basically I am trying to load a model trained using AutoML's Tables with tensorflow. Once a model is trained, it can be exported to Google Storage going to TEST & USE -> USER YOUR MODEL -> Container. After downloading it locally, the model files look like this in my file system:

    model/
    ├── assets
    │   ├── 14_vocab
    │   ├── 15_vocab
    │   ...
    │   ├── 7_vocab
    │   └── 8_vocab
    ├── saved_model.pb
    └── variables
        ├── variables.data-00000-of-00001
        └── variables.index

I've tried to loading the model in several ways

Attempt 1

Using tensorflow==2.0, I tried importing the model like so:

   import tensorflow as tf

   loaded = tf.saved_model.load("./model/")

Which gives me the following error:

    tensorflow.python.framework.errors_impl.NotFoundError: Op type not registered 'ParseExampleV2' in binary running on hostname. 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.

Attempt 2

Using tensorflow==1.13, tensorflow==1.14 and tensorflow==1.15, I tried the same with the following snippet:

   import tensorflow as tf

   with tf.Session(graph=tf.Graph()) as sess:
       tf.saved_model.loader.load(
           sess, [tf.saved_model.tag_constants.SERVING], "./model/"
       )

   # Also tried it this way
   from tensorflow.contrib import predictor

   loaded = predictor.from_saved_model("./model/", signature_def_key='predict')

Which raises a similar error:

    ---------------------------------------------------------------------------
    KeyError                                  Traceback (most recent call last)
    <ipython-input-69-d51c805f9e16> in <module>
          1 with tf.Session(graph=tf.Graph()) as sess:
          2     tf.saved_model.loader.load(
    ----> 3         sess, [tf.saved_model.tag_constants.SERVING], PATH_TO_MODEL
          4     )

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/util/deprecation.py in new_func(*args, **kwargs)
        322               'in a future version' if date is None else ('after %s' % date),
        323               instructions)
    --> 324       return func(*args, **kwargs)
        325     return tf_decorator.make_decorator(
        326         func, new_func, 'deprecated',

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/saved_model/loader_impl.py in load(sess, tags, export_dir, import_scope, **saver_kwargs)
        267   """
        268   loader = SavedModelLoader(export_dir)
    --> 269   return loader.load(sess, tags, import_scope, **saver_kwargs)
        270 
        271 

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/saved_model/loader_impl.py in load(self, sess, tags, import_scope, **saver_kwargs)
        418       `MetagraphDef` proto of the graph that was loaded.
        419     """
    --> 420     with sess.graph.as_default():
        421       saver, _ = self.load_graph(sess.graph, tags, import_scope,
        422                                  **saver_kwargs)

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/saved_model/loader_impl.py in load_graph(self, graph, tags, import_scope, **saver_kwargs)
        348     """
        349     meta_graph_def = self.get_meta_graph_def_from_tags(tags)
    --> 350     with graph.as_default():
        351       return tf_saver._import_meta_graph_with_return_elements(  # pylint: disable=protected-access
        352           meta_graph_def, import_scope=import_scope, **saver_kwargs)

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/training/saver.py in _import_meta_graph_with_return_elements(meta_graph_or_file, clear_devices, import_scope, return_elements, **kwargs)
       1455                                             return_elements=None,
       1456                                             **kwargs):
    -> 1457   """Import MetaGraph, and return both a saver and returned elements."""
       1458   if context.executing_eagerly():
       1459     raise RuntimeError("Exporting/importing meta graphs is not supported when "

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/framework/meta_graph.py in import_scoped_meta_graph_with_return_elements(meta_graph_or_file, clear_devices, graph, import_scope, input_map, unbound_inputs_col_name, restore_collections_predicate, return_elements)
        804       dictionary of all the `Variables` imported into the name scope,
        805       list of `Operation` or `Tensor` objects from the `return_elements` list).
    --> 806 
        807   Raises:
        808     ValueError: If the graph_def contains unbound inputs.

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/util/deprecation.py in new_func(*args, **kwargs)
        505                 'in a future version' if date is None else ('after %s' % date),
        506                 instructions)
    --> 507       return func(*args, **kwargs)
        508 
        509     doc = _add_deprecated_arg_notice_to_docstring(

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/framework/importer.py in import_graph_def(graph_def, input_map, return_elements, name, op_dict, producer_op_list)
        397 
        398   if producer_op_list is not None:
    --> 399     # TODO(skyewm): make a copy of graph_def so we're not mutating the argument?
        400     _RemoveDefaultAttrs(op_dict, producer_op_list, graph_def)
        401 

    ~/anaconda3/envs/tf/lib/python3.7/site-packages/tensorflow/python/framework/importer.py in _RemoveDefaultAttrs(op_dict, producer_op_list, graph_def)
        157   for node in graph_def.node:
        158     # Remove any default attr values that aren't in op_def.
    --> 159     if node.op in producer_op_dict:
        160       op_def = op_dict[node.op]
        161       producer_op_def = producer_op_dict[node.op]

    KeyError: 'ParseExampleV2'

Apparently, there is a missing Op called ParseExampleV2 that's been used by the model created by AutoML's Tables, but I couldn't find a way of loading it.

There is a file in tensorflow's GitHub repository where this Op is mentioned: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/util/example_parser_configuration.py

I've tried importing tensorflow.python.util but it doesn't seem to load this Op... does anyone know why this could be happening?

Answer 1

Sorry about the inconvenience. The model exported from AutoML Tables can only run with the provided model server container.

For how to run the model server container, please follow the tutorial here: https://cloud.google.com/automl-tables/docs/model-export