def test_serialize_deserialize2(self):
td = self.td
td = td.map(lambda x, y: x)
batch = next(iter(td))
td = td.map(serialize_to_example)
td = td.map(make_dataset_deserialize_fn(td))
batch_d = next(iter(td))
self.assertAllEqual(batch, batch_d)
def test_serialize_deserialize0(self):
td = self.td
batch = next(iter(td))
td = td.map(serialize_to_example)
td = td.map(make_dataset_deserialize_fn(td))
batch_d = next(iter(td))
self.assertAllEqual(batch[0], batch_d[0])
self.assertAllEqual(batch[1], batch_d[1])
def test_serialize_deserialize1(self):
td = self.td
td = td.map(lambda x, y: (x, tf.cast(x, tf.float32), y))
batch = next(iter(td))
td = td.map(serialize_to_example)
td = td.map(make_dataset_deserialize_fn(td))
batch_d = next(iter(td))
self.assertAllEqual(batch[0], batch_d[0])
self.assertAllEqual(batch[1], batch_d[1])
self.assertAllEqual(batch[2], batch_d[2])
def test_serialize_deserialize_shape3(self):
td = self.td
batch = next(iter(td))
td = td.map(serialize_to_example)
td = td.map(
make_dataset_deserialize_fn(td, set_shape=True,
set_dimension=True))
self.assertEqual(td.element_spec[0],
tf.TensorSpec(shape=(28, 28, 3), dtype=tf.uint8))
self.assertEqual(td.element_spec[1],
tf.TensorSpec(shape=(), dtype=tf.int64, name=None))
batch_d = next(iter(td))
self.assertAllEqual(batch[0], batch_d[0])
self.assertAllEqual(batch[1], batch_d[1])
def test_serialize_deserialize_variable_shape0(self):
td = self.td
td = td.map(random_size)
batch = next(iter(td))
td = td.map(serialize_to_example)
td = td.map(
make_dataset_deserialize_fn(td,
set_shape=False,
set_dimension=False))
self.assertEqual(td.element_spec[0],
tf.TensorSpec(shape=None, dtype=tf.float32))
self.assertEqual(td.element_spec[1],
tf.TensorSpec(shape=None, dtype=tf.int64, name=None))
batch_d = next(iter(td))
self.assertAllEqual(batch[0], batch_d[0])
self.assertAllEqual(batch[1], batch_d[1])
def test_serialize_deserialize_variable_shape2(self):
td = self.td
s0 = tf.cast(
tf.random.uniform([16, 16, 3], maxval=255, dtype=tf.int32),
tf.uint8)
s1 = tf.random.uniform((), maxval=255, dtype=tf.int64)
td_v = tf.data.Dataset.from_tensors((s0, s1))
td = td_v.concatenate(td)
td = td.map(serialize_to_example)
td = td.map(
make_dataset_deserialize_fn(td,
set_shape=True,
set_dimension=False))
it = iter(td)
x, y = next(it)
with pytest.raises(InvalidArgumentError):
x, y = next(it)