def test_roundtrip_sequence_of_scalars(self):
x = tf.data.Dataset.range(5).map(lambda x: x * 2)
serialized_bytes = tensorflow_serialization.serialize_dataset(x)
y = tensorflow_serialization.deserialize_dataset(serialized_bytes)
self.assertEqual(x.element_spec, y.element_spec)
self.assertAllEqual([y_val for y_val in y], [x * 2 for x in range(5)])
def deserialize_sequence_value(sequence_value_proto):
"""Deserializes a `tf.data.Dataset`.
Args:
sequence_value_proto: `Sequence` protocol buffer message.
Returns:
A tuple of `(tf.data.Dataset, tff.Type)`.
"""
py_typecheck.check_type(sequence_value_proto, executor_pb2.Value.Sequence)
which_value = sequence_value_proto.WhichOneof('value')
if which_value == 'zipped_saved_model':
ds = tensorflow_serialization.deserialize_dataset(
sequence_value_proto.zipped_saved_model)
else:
raise NotImplementedError(
'Deserializing Sequences enocded as {!s} has not been implemented'.
format(which_value))
element_type = type_serialization.deserialize_type(
sequence_value_proto.element_type)
# If a serialized dataset had elements of nested structes of tensors (e.g.
# `dict`, `OrderedDict`), the deserialized dataset will return `dict`,
# `tuple`, or `namedtuple` (loses `collections.OrderedDict` in a conversion).
#
# Since the dataset will only be used inside TFF, we wrap the dictionary
# coming from TF in an `OrderedDict` when necessary (a type that both TF and
# TFF understand), using the field order stored in the TFF type stored during
# serialization.
ds = tensorflow_utils.coerce_dataset_elements_to_tff_type_spec(
ds, element_type)
return ds, computation_types.SequenceType(element=element_type)
def test_roundtrip_sequence_of_scalars(self):
x = tf.data.Dataset.range(5).map(lambda x: x * 2)
serialized_bytes = tensorflow_serialization.serialize_dataset(x)
y = tensorflow_serialization.deserialize_dataset(serialized_bytes)
self.assertEqual(tf.data.experimental.get_structure(x),
tf.data.experimental.get_structure(y))
self.assertAllEqual([y_val for y_val in y], [x * 2 for x in range(5)])
def test_roundtrip_sequence_of_singleton_tuples(self):
x = tf.data.Dataset.range(5).map(lambda x: (x, ))
serialized_bytes = tensorflow_serialization.serialize_dataset(x)
y = tensorflow_serialization.deserialize_dataset(serialized_bytes)
self.assertEqual(x.element_spec, y.element_spec)
expected_values = [(x, ) for x in range(5)]
actual_values = self.evaluate([y_val for y_val in y])
self.assertAllEqual(expected_values, actual_values)
def test_roundtrip_sequence_of_tuples(self):
x = tf.data.Dataset.range(5).map(lambda x: (x * 2, tf.cast(
x, tf.int32), tf.cast(x - 1, tf.float32)))
serialized_bytes = tensorflow_serialization.serialize_dataset(x)
y = tensorflow_serialization.deserialize_dataset(serialized_bytes)
self.assertEqual(x.element_spec, y.element_spec)
self.assertAllEqual(self.evaluate([y_val for y_val in y]),
[(x * 2, x, x - 1.) for x in range(5)])
def test_roundtrip_sequence_of_nested_structures(self):
test_tuple_type = collections.namedtuple('TestTuple', ['u', 'v'])
def _make_nested_tf_structure(x):
return collections.OrderedDict([
('b', tf.cast(x, tf.int32)),
('a',
tuple([
x,
test_tuple_type(x * 2, x * 3),
collections.OrderedDict([('x', x**2), ('y', x**3)])
])),
])
x = tf.data.Dataset.range(5).map(_make_nested_tf_structure)
serialzied_bytes = tensorflow_serialization.serialize_dataset(x)
y = tensorflow_serialization.deserialize_dataset(serialzied_bytes)
# NOTE: TF loses the `OrderedDict` during serialization, so the expectation
# here is for a `dict` in the result.
self.assertEqual(
tf.data.experimental.get_structure(y), {
'b':
tf.TensorSpec([], tf.int32),
'a':
tuple([
tf.TensorSpec([], tf.int64),
test_tuple_type(
tf.TensorSpec([], tf.int64),
tf.TensorSpec([], tf.int64),
),
{
'x': tf.TensorSpec([], tf.int64),
'y': tf.TensorSpec([], tf.int64),
},
]),
})
def _build_expected_structure(x):
return {
'b':
x,
'a':
tuple(
[x,
test_tuple_type(x * 2, x * 3), {
'x': x**2,
'y': x**3
}])
}
actual_values = self.evaluate([y_val for y_val in y])
expected_values = [_build_expected_structure(x) for x in range(5)]
for actual, expected in zip(actual_values, expected_values):
self.assertAllClose(actual, expected)
def test_roundtrip_sequence_of_namedtuples(self):
test_tuple_type = collections.namedtuple('TestTuple', ['a', 'b', 'c'])
def make_test_tuple(x):
return test_tuple_type(
a=x * 2, b=tf.cast(x, tf.int32), c=tf.cast(x - 1, tf.float32))
x = tf.data.Dataset.range(5).map(make_test_tuple)
serialized_bytes = tensorflow_serialization.serialize_dataset(x)
y = tensorflow_serialization.deserialize_dataset(serialized_bytes)
self.assertEqual(x.element_spec, y.element_spec)
self.assertAllEqual(
self.evaluate([y_val for y_val in y]),
[test_tuple_type(a=x * 2, b=x, c=x - 1.) for x in range(5)])
