def testTensorArraySpec(self):
"""Tests tf.TensorArray behavior with `TensorArraySpec` as input signature.
Bugs: b/162452468, b/187114287
Status: Broken
Issue: Using `tf.TensorArraySpec` as the input signature to tf.function
does not work. This is not documented anywhere.
Error message:
"If shallow structure is a sequence, input must also be a sequence."
Notes:
* Documentation for `tf.TensorArraySpec` appears to be minimal. Need to
update it.
"""
self.skipTest('b/187114287')
input_signature = [
tf.TensorArraySpec(
element_shape=None, dtype=tf.float32, dynamic_size=True)
]
@tf.function(input_signature=input_signature)
def f(ta):
return ta.stack()
ta = tf.TensorArray(dtype=tf.float32, dynamic_size=True, size=0)
ta = ta.write(0, tf.constant([1.0, 2.0]))
ta = ta.write(1, tf.constant([3.0, 4.0]))
out_t = tf.constant([[1.0, 2.0], [3.0, 4.0]])
self.assertAllEqual(f(ta), out_t)
def testFailureParamAsDict(self):
"""Tests passing in a `dict` for `failure` param to `_generic_test`."""
def f(ta):
return ta.stack()
ta = tf.TensorArray(dtype=tf.float32, dynamic_size=True, size=0)
ta = ta.write(0, tf.constant([1.0, 2.0]))
ta = ta.write(1, tf.constant([3.0, 4.0]))
out_t = tf.constant([[1.0, 2.0], [3.0, 4.0]])
input_signature = [
tf.TensorArraySpec(element_shape=None,
dtype=tf.float32,
dynamic_size=True)
]
self._generic_test(f, [
Example(arg=(ta, ),
out=out_t,
failure={
RunMode.FUNCTION:
'If shallow structure is a sequence, input must also '
'be a sequence',
RunMode.XLA:
'If shallow structure is a sequence, input must also '
'be a sequence',
RunMode.SAVED:
'Found zero restored functions for caller function',
},
bugs=['b/162452468'])
],
input_signature=input_signature,
skip_modes=[])
return
def _convert_legacy_structure(output_types, output_shapes, output_classes):
"""convert_legacy_structure"""
flat_types = nest.flatten(output_types)
flat_shapes = nest.flatten(output_shapes)
flat_classes = nest.flatten(output_classes)
flat_ret = []
for flat_type, flat_shape, flat_class in zip(flat_types, flat_shapes,
flat_classes):
if isinstance(flat_class, tf.TypeSpec):
flat_ret.append(flat_class)
elif issubclass(flat_class, tf.sparse.SparseTensor):
flat_ret.append(tf.SparseTensorSpec(flat_shape, flat_type))
elif issubclass(flat_class, tf.Tensor):
flat_ret.append(tf.TensorSpec(flat_shape, flat_type))
elif issubclass(flat_class, tf.TensorArray):
# We sneaked the dynamic_size and infer_shape into the legacy shape.
flat_ret.append(
tf.TensorArraySpec(
flat_shape[2:],
flat_type,
dynamic_size=tf.compat.dimension_value(flat_shape[0]),
infer_shape=tf.compat.dimension_value(flat_shape[1]),
))
else:
# NOTE(mrry): Since legacy structures produced by iterators only
# comprise Tensors, SparseTensors, and nests, we do not need to
# support all structure types here.
raise TypeError(
"Could not build a structure for output class {!r}".format(
flat_class))
return nest.pack_sequence_as(output_classes, flat_ret)