def test_dict_to_tf_example_error_reraise(self):
# Test error reraise in _dict_to_tf_example.
example_data = {'input': [1, 2, 3]}
tensor_info = {
'input': feature_lib.TensorInfo(
shape=(2, ),
dtype=tf.int64,
),
}
with self.assertRaisesRegex(
ValueError, 'Error while serializing feature `input`:'):
example_serializer._dict_to_tf_example(example_data, tensor_info)
def test_ragged_dict_to_tf_example_empty(self):
example_data = {
'input': [],
}
tensor_info = {
'input':
feature_lib.TensorInfo(
shape=(
None,
None,
),
dtype=tf.int64,
sequence_rank=2,
),
}
ex_proto = example_serializer._dict_to_tf_example(
example_data, tensor_info)
feature = ex_proto.features.feature
self.assertEqual(
[],
list(feature['input/ragged_flat_values'].int64_list.value),
)
self.assertEqual(
[],
list(feature['input/ragged_row_lengths_0'].int64_list.value),
)
def test_dict_to_example(self):
example = example_serializer._dict_to_tf_example({
"a":
1,
"a2":
np.array(1),
"b": ["foo", "bar"],
"b2":
np.array(["foo", "bar"]),
"c": [2.0],
"c2":
np.array([2.0]),
# Empty values supported when type is defined
"d":
np.array([], dtype=np.int32),
# Support for byte strings
"e":
np.zeros(2, dtype=np.uint8).tobytes(),
"e2": [np.zeros(2, dtype=np.uint8).tobytes()] * 2,
})
feature = example.features.feature
self.assertEqual([1], list(feature["a"].int64_list.value))
self.assertEqual([1], list(feature["a2"].int64_list.value))
self.assertEqual([b"foo", b"bar"], list(feature["b"].bytes_list.value))
self.assertEqual([b"foo", b"bar"],
list(feature["b2"].bytes_list.value))
self.assertEqual([2.0], list(feature["c"].float_list.value))
self.assertEqual([2.0], list(feature["c2"].float_list.value))
self.assertEqual([], list(feature["d"].int64_list.value))
self.assertEqual([b"\x00\x00"], list(feature["e"].bytes_list.value))
self.assertEqual([b"\x00\x00", b"\x00\x00"],
list(feature["e2"].bytes_list.value))
with self.assertRaisesWithPredicateMatch(ValueError,
"Received an empty"):
# Raise error if an undefined empty value is given
example_serializer._dict_to_tf_example({
"empty": [],
})
with self.assertRaisesWithPredicateMatch(ValueError,
"not support type"):
# Raise error if an unsupported dtype is given
example_serializer._dict_to_tf_example({
"wrong_type":
np.zeros(shape=(5, ), dtype=np.complex64),
})
