Using tf.data.Dataset with tf Hub Modules - tensorflow2.0

Question

How do I feed a tf.keras model, that includes a 1D input TF Hub module, with a tf.data.Dataset?
(Ultimately, the aim is to use a single tf.data.Dataset with a multi-input, multi-output keras funtional api model.)
Tried this:
import tensorflow as tf
import tensorflow_hub as hub
embed = "https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim/1"
hub_layer = hub.KerasLayer(embed, output_shape=[20], input_shape=[],
dtype=tf.string, trainable=True, name='hub_layer')
# From tf hub webpage: "The module takes a batch of sentences in a 1-D tensor of strings as input."
input_tensor = tf.keras.Input(shape=(), dtype=tf.string)
hub_tensor = hub_layer(input_tensor)
x = tf.keras.layers.Dense(16, activation='relu')(hub_tensor)#(x)
main_output = tf.keras.layers.Dense(units=4, activation='softmax', name='main_output')(x)
model = tf.keras.models.Model(inputs=[input_tensor], outputs=[main_output])
# This works as expected.
X_tensor = tf.constant(['Hello World', 'The Quick Brown Fox'])
model(X_tensor)
# This fails
X_ds = tf.data.Dataset.from_tensors(X_tensor)
X_ds.element_spec
model(X_ds)
Expectation was that the 1D tensor in the dataset would be automatically extracted and consumed by the model.
Error message:
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
in
21 X_ds = tf.data.Dataset.from_tensors(X_tensor)
22 X_ds.element_spec
---> 23 model(X_ds)
24
25
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self, *args, **kwargs)
966 with base_layer_utils.autocast_context_manager(
967 self._compute_dtype):
--> 968 outputs = self.call(cast_inputs, *args, **kwargs)
969 self._handle_activity_regularization(inputs, outputs)
970 self._set_mask_metadata(inputs, outputs, input_masks)
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py in call(self, inputs, training, mask)
717 return self._run_internal_graph(
718 inputs, training=training, mask=mask,
--> 719 convert_kwargs_to_constants=base_layer_utils.call_context().saving)
720
721 def compute_output_shape(self, input_shape):
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py in _run_internal_graph(self, inputs, training, mask, convert_kwargs_to_constants)
835 tensor_dict = {}
836 for x, y in zip(self.inputs, inputs):
--> 837 y = self._conform_to_reference_input(y, ref_input=x)
838 x_id = str(id(x))
839 tensor_dict[x_id] = [y] * self._tensor_usage_count[x_id]
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py in _conform_to_reference_input(self, tensor, ref_input)
959 # Dtype handling.
960 if isinstance(ref_input, (ops.Tensor, composite_tensor.CompositeTensor)):
--> 961 tensor = math_ops.cast(tensor, dtype=ref_input.dtype)
962
963 return tensor
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/util/dispatch.py in wrapper(*args, **kwargs)
178 """Call target, and fall back on dispatchers if there is a TypeError."""
179 try:
--> 180 return target(*args, **kwargs)
181 except (TypeError, ValueError):
182 # Note: convert_to_eager_tensor currently raises a ValueError, not a
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py in cast(x, dtype, name)
785 # allows some conversions that cast() can't do, e.g. casting numbers to
786 # strings.
--> 787 x = ops.convert_to_tensor(x, name="x")
788 if x.dtype.base_dtype != base_type:
789 x = gen_math_ops.cast(x, base_type, name=name)
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/framework/ops.py in convert_to_tensor(value, dtype, name, as_ref, preferred_dtype, dtype_hint, ctx, accepted_result_types)
1339
1340 if ret is None:
-> 1341 ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
1342
1343 if ret is NotImplemented:
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py in _constant_tensor_conversion_function(v, dtype, name, as_ref)
319 as_ref=False):
320 _ = as_ref
--> 321 return constant(v, dtype=dtype, name=name)
322
323
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py in constant(value, dtype, shape, name)
260 """
261 return _constant_impl(value, dtype, shape, name, verify_shape=False,
--> 262 allow_broadcast=True)
263
264
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py in _constant_impl(value, dtype, shape, name, verify_shape, allow_broadcast)
268 ctx = context.context()
269 if ctx.executing_eagerly():
--> 270 t = convert_to_eager_tensor(value, ctx, dtype)
271 if shape is None:
272 return t
~/projects/email_analysis/email_venv/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
94 dtype = dtypes.as_dtype(dtype).as_datatype_enum
95 ctx.ensure_initialized()
---> 96 return ops.EagerTensor(value, ctx.device_name, dtype)
97
98
ValueError: Attempt to convert a value () with an unsupported type () to a Tensor.

Answer 1

The point of a dataset is to provide a sequence of tensors, like here:
all_data = tf.constant([['Hello', 'World'], ['Brown Fox', 'lazy dog']])
ds = tf.data.Dataset.from_tensor_slices(all_data)
for tensor in ds:
print(tensor)
which outputs
tf.Tensor([b'Hello' b'World'], shape=(2,), dtype=string)
tf.Tensor([b'Brown Fox' b'lazy dog'], shape=(2,), dtype=string)
Instead of just printing tensor, you can compute with it:
for tensor in ds:
print(hub_layer(tensor))
which outputs 2 tensors of shape (2,20) each.
For more, see https://www.tensorflow.org/guide/data.