shape of example features don't match expectation when using example keras code

Hello,

I’m trying to run the example code using the keras model and the tf_record data. The problem is that the shape of the example feature is (None, None, None) but expected to have only 2 dimensions.

It is probably something stupid. Hope somebody can point me towards the error.

Here is my example code to recreate the problem. It is mostly copy paste from the examples. (error msg is below)

from google.protobuf import text_format
from tensorflow_serving.apis import input_pb2
import os
import tensorflow as tf
import tensorflow_ranking as tfr


_FILE_NAME = "/tmp/ranking_example.tf_record"
_LABEL_FEATURE = "relevance"
_PADDING_LABEL = -1
_SIZE="example_list_size"


def create_feature_columns():
  sparse_column = tf.feature_column.categorical_column_with_hash_bucket(
      key="user_id", hash_bucket_size=100, dtype=tf.int64)
  query_embedding = tf.feature_column.embedding_column(
      categorical_column=sparse_column, dimension=20)
  context_feature_columns = {"user_id": query_embedding}

  sparse_column = tf.feature_column.categorical_column_with_hash_bucket(
      key="document_id", hash_bucket_size=100, dtype=tf.int64)
  document_embedding = tf.feature_column.embedding_column(
      categorical_column=sparse_column, dimension=20)
  example_feature_columns = {"document_id": document_embedding}

  return context_feature_columns, example_feature_columns


def make_dataset(file_pattern,
                 batch_size,
                 randomize_input=False,
                 num_epochs=None):
  context_feature_columns, example_feature_columns = create_feature_columns()
  context_feature_spec = tf.feature_column.make_parse_example_spec(
      context_feature_columns.values())
  label_column = tf.feature_column.numeric_column(
      _LABEL_FEATURE, dtype=tf.int64, default_value=_PADDING_LABEL)
  example_feature_spec = tf.feature_column.make_parse_example_spec(
      list(example_feature_columns.values()) + [label_column])
  dataset = tfr.data.build_ranking_dataset(
      file_pattern=file_pattern,
      data_format=tfr.data.ELWC,
      batch_size=batch_size,
      context_feature_spec=context_feature_spec,
      example_feature_spec=example_feature_spec,
      reader=tf.data.TFRecordDataset,
      shuffle=randomize_input,
      num_epochs=num_epochs,
      size_feature_name=_SIZE)

  def _separate_features_and_label(features):
    label = tf.squeeze(features.pop(_LABEL_FEATURE), axis=2)
    label = tf.cast(label, tf.float32)
    return features, label

  dataset = dataset.map(_separate_features_and_label)
  return dataset


def test_ranking_example():
    samples = text_format.Parse(
        """
        context {
          features {
            feature {
              key: "user_id"
              value { int64_list { value: 1 } }
            }
          }
        }
        examples {
          features {
            feature {
              key: "document_id"
              value { int64_list { value: 1 } }
            }
            feature {
              key: "relevance"
              value { int64_list { value: 1 } }
            }
          }
        }
        examples {
          features {
            feature {
              key: "document_id"
              value { int64_list { value: 2 } } 
            }
            feature {
              key: "relevance"
              value { int64_list { value: 0 } }
            }
          }
        }""", input_pb2.ExampleListWithContext())

    try:
        os.remove(_FILE_NAME)
    except FileNotFoundError:
        pass

    with tf.io.TFRecordWriter(_FILE_NAME) as writer:
        for sample in [samples]*6:
            writer.write(sample.SerializeToString())

    batch_size = 2
    dataset = make_dataset(_FILE_NAME, batch_size)
    context_feature_columns, example_feature_columns = create_feature_columns()
    # Use a Premade Network, or subclass and build your own!
    network = tfr.keras.canned.DNNRankingNetwork(
        context_feature_columns=context_feature_columns,
        example_feature_columns=example_feature_columns,
        hidden_layer_dims=[1024, 512, 256],
        activation=tf.nn.relu,
        dropout=0.5)

    softmax_loss_obj = tfr.keras.losses.get(tfr.losses.RankingLossKey.SOFTMAX_LOSS)

    # Contains all ranking metrics, including NDCG @ {1, 3, 5, 10}.
    default_metrics = tfr.keras.metrics.default_keras_metrics()

    ranker = tfr.keras.model.create_keras_model(
        network=network,
        loss=softmax_loss_obj,
        metrics=default_metrics,
        optimizer=tf.keras.optimizers.Adagrad(learning_rate=0.05),
        size_feature_name=_SIZE)

    r = ranker.fit(
        dataset,
        steps_per_epoch=4,
        epochs=10
    )

    assert r

error:

test_ranking_all.py:132: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py:819: in fit
    use_multiprocessing=use_multiprocessing)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py:235: in fit
    use_multiprocessing=use_multiprocessing)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py:593: in _process_training_inputs
    use_multiprocessing=use_multiprocessing)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py:706: in _process_inputs
    use_multiprocessing=use_multiprocessing)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/data_adapter.py:702: in __init__
    x = standardize_function(x)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py:684: in standardize_function
    return dataset.map(map_fn, num_parallel_calls=dataset_ops.AUTOTUNE)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/data/ops/dataset_ops.py:1591: in map
    self, map_func, num_parallel_calls, preserve_cardinality=True)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/data/ops/dataset_ops.py:3926: in __init__
    use_legacy_function=use_legacy_function)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/data/ops/dataset_ops.py:3147: in __init__
    self._function = wrapper_fn._get_concrete_function_internal()
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py:2395: in _get_concrete_function_internal
    *args, **kwargs)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py:2389: in _get_concrete_function_internal_garbage_collected
    graph_function, _, _ = self._maybe_define_function(args, kwargs)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py:2703: in _maybe_define_function
    graph_function = self._create_graph_function(args, kwargs)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py:2593: in _create_graph_function
    capture_by_value=self._capture_by_value),
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/framework/func_graph.py:978: in func_graph_from_py_func
    func_outputs = python_func(*func_args, **func_kwargs)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/data/ops/dataset_ops.py:3140: in wrapper_fn
    ret = _wrapper_helper(*args)
../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/data/ops/dataset_ops.py:3082: in _wrapper_helper
    ret = autograph.tf_convert(func, ag_ctx)(*nested_args)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = ({'document_id': <tensorflow.python.framework.sparse_tensor.SparseTensor object at 0x7f47446fded0>, 'example_list_size....framework.sparse_tensor.SparseTensor object at 0x7f4744716590>}, <tf.Tensor 'args_3:0' shape=(2, None) dtype=float32>)
kwargs = {}
options = <tensorflow.python.autograph.core.converter.ConversionOptions object at 0x7f4744716950>

    def wrapper(*args, **kwargs):
      """Wrapper that calls the converted version of f."""
      options = converter.ConversionOptions(
          recursive=recursive,
          user_requested=user_requested,
          optional_features=optional_features)
      try:
        return converted_call(f, args, kwargs, options=options)
      except Exception as e:  # pylint:disable=broad-except
        if hasattr(e, 'ag_error_metadata'):
>         raise e.ag_error_metadata.to_exception(e)
E         ValueError: in converted code:
E         
E             /content-recommendations/.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py:677 map_fn
E                 batch_size=None)
E             /content-recommendations/.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py:2410 _standardize_tensors
E                 exception_prefix='input')
E             content-recommendations/.venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_utils.py:573 standardize_input_data
E                 'with shape ' + str(data_shape))
E         
E             ValueError: Error when checking input: expected document_id to have 2 dimensions, but got array with shape (None, None, None)

../../../.venv/lib/python3.7/site-packages/tensorflow_core/python/autograph/impl/api.py:237: ValueError

pip list:

Package                Version    
---------------------- -----------
absl-py                0.9.0      
alembic                1.4.2      
aniso8601              7.0.0      
appdirs                1.4.3      
astor                  0.8.1      
attrs                  19.3.0     
black                  19.10b0    
bleach                 3.1.4      
boto3                  1.9.253    
botocore               1.12.253   
cached-property        1.5.1      
cachetools             4.1.0      
certifi                2020.4.5.1 
cfgv                   3.1.0      
chardet                3.0.4      
click                  7.1.1      
coloredlogs            14.0       
contextlib2            0.6.0.post1
cycler                 0.10.0     
dagit                  0.7.7      
dagster                0.7.7      
dagster-aws            0.7.7      
dagster-cron           0.7.7      
dagster-graphql        0.7.7      
dagster-pandas         0.7.7      
decorator              4.4.2      
defusedxml             0.6.0      
distlib                0.3.0      
docutils               0.15.2     
entrypoints            0.3        
filelock               3.0.12     
Flask                  1.1.2      
Flask-Cors             3.0.8      
Flask-GraphQL          2.0.1      
Flask-Sockets          0.2.1      
funcsigs               1.0.2      
future                 0.18.2     
gast                   0.2.2      
gevent                 20.4.0     
gevent-websocket       0.10.1     
google-auth            1.14.0     
google-auth-oauthlib   0.4.1      
google-pasta           0.2.0      
graphene               2.1.8      
graphql-core           2.3.1      
graphql-relay          2.0.1      
graphql-server-core    1.2.0      
graphql-ws             0.3.0      
graphviz               0.14       
greenlet               0.4.15     
grpcio                 1.28.1     
h5py                   2.10.0     
humanfriendly          8.2        
identify               1.4.15     
idna                   2.9        
importlib-metadata     1.6.0      
ipython-genutils       0.2.0      
itsdangerous           1.1.0      
Jinja2                 2.11.2     
jmespath               0.9.5      
jsonschema             3.2.0      
jupyter-core           4.6.3      
Keras-Applications     1.0.8      
Keras-Preprocessing    1.1.0      
kiwisolver             1.2.0      
Mako                   1.1.2      
Markdown               3.2.1      
MarkupSafe             1.1.1      
matplotlib             3.2.1      
mistune                0.8.4      
more-itertools         8.2.0      
nbconvert              5.6.1      
nbformat               5.0.6      
nodeenv                1.3.5      
numpy                  1.18.3     
oauthlib               3.1.0      
opt-einsum             3.2.1      
packaging              20.3       
pandas                 1.0.3      
pandocfilters          1.4.2      
pathspec               0.8.0      
pathtools              0.1.2      
pip                    20.0.2     
pip-tools              5.0.0      
pkg-resources          0.0.0      
pluggy                 0.13.1     
pre-commit             2.3.0      
promise                2.3        
protobuf               3.11.3     
psycopg2-binary        2.8.5      
py                     1.8.1      
pyarrow                0.17.0     
pyasn1                 0.4.8      
pyasn1-modules         0.2.8      
Pygments               2.6.1      
PyMySQL                0.9.3      
pyparsing              2.4.7      
pyrsistent             0.16.0     
pytest                 5.4.1      
python-crontab         2.4.1      
python-dateutil        2.8.1      
python-editor          1.0.4      
pytz                   2019.3     
PyYAML                 5.3.1      
regex                  2020.4.4   
requests               2.23.0     
requests-oauthlib      1.3.0      
rsa                    4.0        
Rx                     1.6.1      
s3transfer             0.2.1      
scipy                  1.4.1      
setuptools             46.1.3     
six                    1.14.0     
SQLAlchemy             1.3.16     
tensorboard            2.1.1      
tensorflow             2.1.0      
tensorflow-estimator   2.1.0      
tensorflow-ranking     0.3.0      
tensorflow-serving-api 2.1.0      
termcolor              1.1.0      
terminaltables         3.1.0      
testpath               0.4.4      
toml                   0.10.0     
toposort               1.5        
tqdm                   4.45.0     
traitlets              4.3.3      
typed-ast              1.4.1      
urllib3                1.25.9     
virtualenv             20.0.18    
watchdog               0.10.2     
wcwidth                0.1.9      
webencodings           0.5.1      
Werkzeug               1.0.1      
wheel                  0.34.2     
wrapt                  1.12.1     
zipp                   3.1.0  

[filthysocks]

@filthysocks curious: why did you choose making your document_id as a categorical_column_with_hash_bucket column VS numeric column ?

well since its categorical data and tensorflow only offers here three different solutions (see https://www.tensorflow.org/tutorials/structured_data/feature_columns#categorical_columns). A numeric column wouldn’t make much sense it would imply that there is somehow a meaning in the id value itself (e.g. 3x the ID = 3x better).

I came across the same error trying to fit the keras model on my own data. I dug around and traced the bug to this line of code https://github.com/tensorflow/ranking/blob/master/tensorflow_ranking/python/keras/feature.py#L61 shape = [list_size] + list(spec.shape) if not is_sparse else (1,). I changed my local source to shape = [list_size] + list(spec.shape) if not is_sparse else (list_size, None) and it ran. @ramakumar1729 Looking forward to your validation and possibly a quick bug fix release. Thx!

cool. hope it’ll work

[ramakumar1729]

I came across the same error trying to fit the keras model on my own data. I dug around and traced the bug to this line of code https://github.com/tensorflow/ranking/blob/master/tensorflow_ranking/python/keras/feature.py#L61 shape = [list_size] + list(spec.shape) if not is_sparse else (1,). I changed my local source to shape = [list_size] + list(spec.shape) if not is_sparse else (list_size, None) and it ran. @ramakumar1729 Looking forward to your validation and possibly a quick bug fix release. Thx!

@yzhangswingman : Thanks for finding the root cause of the issue! This needs fixing, and I’ll update it in the next release.