def testAddTensorListsFailsIfLeadingDimsMismatch(self):
with self.cached_session(), self.test_scope():
l1 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=2)
l2 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=3)
l = math_ops.add_n([l1, l2])
with self.assertRaisesRegexp(
errors.InvalidArgumentError,
"TensorList arguments to AddN must all have the same shape"):
list_ops.tensor_list_stack(l, element_dtype=dtypes.float32).eval()
def testAddTensorListsFailsIfLeadingDimsMismatch(self):
with self.session(), self.test_scope():
l1 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=2)
l2 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=3)
l = math_ops.add_n([l1, l2])
with self.assertRaisesRegex(
errors.InvalidArgumentError,
"TensorList arguments to AddN must all have the same shape"):
list_ops.tensor_list_stack(l, element_dtype=dtypes.float32).eval()
def testAddTensorLists(self):
with self.session(), self.test_scope():
l1 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=3)
l2 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=3)
l1 = list_ops.tensor_list_set_item(l1, 0, 5.)
l2 = list_ops.tensor_list_set_item(l2, 2, 10.)
l = math_ops.add_n([l1, l2])
self.assertAllEqual(
list_ops.tensor_list_stack(l, element_dtype=dtypes.float32),
[5.0, 0.0, 10.0])
def testAddTensorLists(self):
with self.cached_session(), self.test_scope():
l1 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=3)
l2 = list_ops.tensor_list_reserve(
element_shape=[], element_dtype=dtypes.float32, num_elements=3)
l1 = list_ops.tensor_list_set_item(l1, 0, 5.)
l2 = list_ops.tensor_list_set_item(l2, 2, 10.)
l = math_ops.add_n([l1, l2])
self.assertAllEqual(
list_ops.tensor_list_stack(l, element_dtype=dtypes.float32),
[5.0, 0.0, 10.0])
def testStackWithUninitializedTensors(self):
with self.cached_session(), self.test_scope():
l = list_ops.tensor_list_reserve(element_dtype=dtypes.float32,
element_shape=[],
num_elements=3)
t = list_ops.tensor_list_stack(l, element_dtype=dtypes.float32)
self.assertAllEqual(t, [0., 0., 0.])
def build_graph(parameters):
"""Build the TensorListSetItem op testing graph."""
item = tf.placeholder(dtype=parameters["element_dtype"],
shape=parameters["element_shape"])
tensor_list = list_ops.tensor_list_reserve(
element_shape=None,
num_elements=parameters["num_elements"],
element_dtype=parameters["element_dtype"])
init_state = (0, tensor_list)
condition = lambda i, _: i < parameters["num_elements"]
def loop_body(i, tensor_list):
new_item = tf.add(
tf.add(item, item),
tf.constant(value=1, dtype=parameters["element_dtype"]))
new_list = list_ops.tensor_list_set_item(tensor_list, i, new_item)
return i + 1, new_list
_, tensor_list = tf.while_loop(condition, loop_body, init_state)
out = list_ops.tensor_list_stack(
tensor_list,
num_elements=parameters["num_elements"],
element_dtype=parameters["element_dtype"])
return [item], [out]
def test_numpyFailsForVariant(self):
variant_t = list_ops.tensor_list_reserve(element_shape=[],
num_elements=1,
element_dtype=dtypes.float32)
with self.assertRaisesRegex(errors.InvalidArgumentError,
"Cannot convert .+ variant"):
variant_t._numpy()
def testSetStackReservedUnknownElementShape(self):
with self.cached_session(), self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=None, num_elements=2)
l = list_ops.tensor_list_set_item(l, 0, [3.0, 4.0])
t = list_ops.tensor_list_stack(l, element_dtype=dtypes.float32)
self.assertAllEqual(t, [[3.0, 4.0], [0., 0.]])
def testSetStackReservedUnknownElementShape(self):
with self.cached_session(), self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=None, num_elements=2)
l = list_ops.tensor_list_set_item(l, 0, [3.0, 4.0])
t = list_ops.tensor_list_stack(l, element_dtype=dtypes.float32)
self.assertAllEqual(t, [[3.0, 4.0], [0., 0.]])
def testGetSetReserved(self):
with self.cached_session(), self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=2)
e0 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
self.assertAllEqual(e0, 0.0)
l = list_ops.tensor_list_set_item(l, 0, 3.0)
t = list_ops.tensor_list_stack(l, element_dtype=dtypes.float32)
self.assertAllEqual(t, [3.0, 0.0])
def testGetSetReserved(self):
with self.cached_session(), self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=2)
e0 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
self.assertAllEqual(e0, 0.0)
l = list_ops.tensor_list_set_item(l, 0, 3.0)
t = list_ops.tensor_list_stack(l, element_dtype=dtypes.float32)
self.assertAllEqual(t, [3.0, 0.0])
def testSetGetGrad(self):
with backprop.GradientTape() as tape:
t = constant_op.constant(5.)
tape.watch(t)
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=3)
l = list_ops.tensor_list_set_item(l, 1, 2. * t)
e = list_ops.tensor_list_get_item(l, 1, element_dtype=dtypes.float32)
self.assertAllEqual(self.evaluate(e), 10.0)
self.assertAllEqual(self.evaluate(tape.gradient(e, t)), 2.0)
def testSkipEagerTensorListGetItemGradAggregation(self):
l = list_ops.tensor_list_reserve(
element_shape=[], num_elements=1, element_dtype=dtypes.float32)
x = constant_op.constant(1.0)
l = list_ops.tensor_list_set_item(l, 0, x)
l_read1 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
l_read2 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
grad = gradients_impl.gradients([l_read1, l_read2], [x])
with self.cached_session() as sess:
self.assertSequenceEqual(self.evaluate(grad), [2.])
def testSetGetGrad(self):
with backprop.GradientTape() as tape:
t = constant_op.constant(5.)
tape.watch(t)
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=3)
l = list_ops.tensor_list_set_item(l, 1, 2. * t)
e = list_ops.tensor_list_get_item(l, 1, element_dtype=dtypes.float32)
self.assertAllEqual(self.evaluate(e), 10.0)
self.assertAllEqual(self.evaluate(tape.gradient(e, t)), 2.0)
def testSkipEagerTensorListGetItemGradAggregation(self):
l = list_ops.tensor_list_reserve(
element_shape=[], num_elements=1, element_dtype=dtypes.float32)
x = constant_op.constant(1.0)
l = list_ops.tensor_list_set_item(l, 0, x)
l_read1 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
l_read2 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
grad = gradients_impl.gradients([l_read1, l_read2], [x])
with self.cached_session() as sess:
self.assertSequenceEqual(self.evaluate(grad), [2.])
def testElementShape(self):
with self.cached_session() as sess, self.test_scope():
dim = array_ops.placeholder(dtypes.int32)
l = list_ops.tensor_list_reserve(
element_shape=(dim, 15), num_elements=20,
element_dtype=dtypes.float32)
e32 = list_ops.tensor_list_element_shape(l, shape_type=dtypes.int32)
e64 = list_ops.tensor_list_element_shape(l, shape_type=dtypes.int64)
self.assertAllEqual(sess.run(e32, {dim: 10}), (10, 15))
self.assertAllEqual(sess.run(e64, {dim: 7}), (7, 15))
def testElementShape(self):
with self.cached_session() as sess, self.test_scope():
dim = array_ops.placeholder(dtypes.int32)
l = list_ops.tensor_list_reserve(
element_shape=(dim, 15), num_elements=20,
element_dtype=dtypes.float32)
e32 = list_ops.tensor_list_element_shape(l, shape_type=dtypes.int32)
e64 = list_ops.tensor_list_element_shape(l, shape_type=dtypes.int64)
self.assertAllEqual(sess.run(e32, {dim: 10}), (10, 15))
self.assertAllEqual(sess.run(e64, {dim: 7}), (7, 15))
def testGetSetReservedNonScalar(self):
with self.cached_session() as sess, self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32,
element_shape=(7, 15),
num_elements=2)
l = list_ops.tensor_list_set_item(
l, 0, constant_op.constant(1.0, shape=(7, 15)))
e1 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
e2 = list_ops.tensor_list_get_item(l, 1, element_dtype=dtypes.float32)
self.assertAllEqual(sess.run(e1), np.ones((7, 15)))
self.assertAllEqual(sess.run(e2), np.zeros((7, 15)))
def testGetSetReservedNonScalar(self):
with self.cached_session() as sess, self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32,
element_shape=(7, 15),
num_elements=2)
l = list_ops.tensor_list_set_item(
l, 0, constant_op.constant(1.0, shape=(7, 15)))
e1 = list_ops.tensor_list_get_item(l, 0, element_dtype=dtypes.float32)
e2 = list_ops.tensor_list_get_item(l, 1, element_dtype=dtypes.float32)
self.assertAllEqual(sess.run(e1), np.ones((7, 15)))
self.assertAllEqual(sess.run(e2), np.zeros((7, 15)))
def testPushPop(self):
with self.cached_session() as sess, self.test_scope():
l = list_ops.tensor_list_reserve(
element_shape=(7, 15), num_elements=10, element_dtype=dtypes.float32)
l = list_ops.tensor_list_push_back(
l, constant_op.constant(1.0, shape=(7, 15)))
l = list_ops.tensor_list_push_back(
l, constant_op.constant(2.0, shape=(7, 15)))
l, e2 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
_, e1 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
self.assertAllEqual(sess.run(e2), 2.0 * np.ones((7, 15)))
self.assertAllEqual(sess.run(e1), 1.0 * np.ones((7, 15)))
def testPushPop(self):
with self.cached_session() as sess, self.test_scope():
num = array_ops.placeholder(dtypes.int32)
l = list_ops.tensor_list_reserve(
element_shape=(7, 15), num_elements=num, element_dtype=dtypes.float32)
l = list_ops.tensor_list_push_back(
l, constant_op.constant(1.0, shape=(7, 15)))
l = list_ops.tensor_list_push_back(
l, constant_op.constant(2.0, shape=(7, 15)))
l, e2 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
_, e1 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
self.assertAllEqual(sess.run(e2, {num: 10}), 2.0 * np.ones((7, 15)))
self.assertAllEqual(sess.run(e1, {num: 10}), 1.0 * np.ones((7, 15)))
def model(x):
l = list_ops.tensor_list_reserve(element_dtype=tf.int64,
element_shape=[None, 1],
num_elements=2)
init_state = (0, x, l)
condition = lambda i, x, l: i < 2
def body(i, x, l):
element = tf.where(x[i])
l = list_ops.tensor_list_set_item(l, i, element)
return i + 1, x, l
_, _, l_final = tf.while_loop(condition, body, init_state)
return list_ops.tensor_list_stack(l_final, element_dtype=tf.int64)
def testAddTensorListsFailsIfElementShapesMismatch(self):
with self.session() as session, self.test_scope():
# Use placeholders instead of constant values for shapes to prevent TF's
# shape inference from catching this early.
l1_element_shape = array_ops.placeholder(dtype=dtypes.int32)
l2_element_shape = array_ops.placeholder(dtype=dtypes.int32)
l1 = list_ops.tensor_list_reserve(
element_shape=l1_element_shape,
element_dtype=dtypes.float32,
num_elements=3)
l2 = list_ops.tensor_list_reserve(
element_shape=l2_element_shape,
element_dtype=dtypes.float32,
num_elements=3)
l = math_ops.add_n([l1, l2])
with self.assertRaisesRegex(
errors.InvalidArgumentError,
"TensorList arguments to AddN must all have the same shape"):
session.run(
list_ops.tensor_list_stack(l, element_dtype=dtypes.float32), {
l1_element_shape: [],
l2_element_shape: [2]
})
def testSerializeListWithInvalidTensors(self):
worker = test_util.create_local_cluster(num_workers=1, num_ps=1)[0][0]
with ops.Graph().as_default(), session.Session(target=worker.target):
with ops.device("/job:worker"):
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=2)
l = list_ops.tensor_list_set_item(l, 0, 1.)
with ops.device("/job:ps"):
l_ps = array_ops.identity(l)
l_ps = list_ops.tensor_list_set_item(l_ps, 1, 2.)
t = list_ops.tensor_list_stack(l_ps, element_dtype=dtypes.float32)
with ops.device("/job:worker"):
worker_t = array_ops.identity(t)
self.assertAllEqual(self.evaluate(worker_t), [1.0, 2.0])
def testSerializeListWithInvalidTensors(self):
worker = test_util.create_local_cluster(num_workers=1, num_ps=1)[0][0]
with ops.Graph().as_default(), session.Session(target=worker.target):
with ops.device("/job:worker"):
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=2)
l = list_ops.tensor_list_set_item(l, 0, 1.)
with ops.device("/job:ps"):
l_ps = array_ops.identity(l)
l_ps = list_ops.tensor_list_set_item(l_ps, 1, 2.)
t = list_ops.tensor_list_stack(l_ps, element_dtype=dtypes.float32)
with ops.device("/job:worker"):
worker_t = array_ops.identity(t)
self.assertAllEqual(self.evaluate(worker_t), [1.0, 2.0])
def testAddTensorListsFailsIfElementShapesMismatch(self):
with self.cached_session() as session, self.test_scope():
# Use placeholders instead of constant values for shapes to prevent TF's
# shape inference from catching this early.
l1_element_shape = array_ops.placeholder(dtype=dtypes.int32)
l2_element_shape = array_ops.placeholder(dtype=dtypes.int32)
l1 = list_ops.tensor_list_reserve(
element_shape=l1_element_shape,
element_dtype=dtypes.float32,
num_elements=3)
l2 = list_ops.tensor_list_reserve(
element_shape=l2_element_shape,
element_dtype=dtypes.float32,
num_elements=3)
l = math_ops.add_n([l1, l2])
with self.assertRaisesRegexp(
errors.InvalidArgumentError,
"TensorList arguments to AddN must all have the same shape"):
session.run(
list_ops.tensor_list_stack(l, element_dtype=dtypes.float32), {
l1_element_shape: [],
l2_element_shape: [2]
})
def testEagerTensorFormatForVariant(self):
t = list_ops.tensor_list_reserve(element_shape=[1],
num_elements=1,
element_dtype=dtypes.float32)
self.assertEqual(f"{t}", "<TensorList>")
self.assertEqual(str(t),
"tf.Tensor(<TensorList>, shape=(), dtype=variant)")
self.assertEqual(f"{t!s}",
"tf.Tensor(<TensorList>, shape=(), dtype=variant)")
self.assertEqual(
repr(t),
"<tf.Tensor: shape=(), dtype=variant, value=<TensorList>>")
self.assertEqual(
f"{t!r}",
"<tf.Tensor: shape=(), dtype=variant, value=<TensorList>>")
def testZerosLikeUninitialized(self):
l0 = list_ops.tensor_list_reserve([], 3, element_dtype=dtypes.float32)
l1 = list_ops.tensor_list_set_item(l0, 0, 1.) # [1., _, _]
zeros_1 = array_ops.zeros_like(l1) # [0., _, _]
l2 = list_ops.tensor_list_set_item(l1, 2, 2.) # [1., _, 2.]
zeros_2 = array_ops.zeros_like(l2) # [0., _, 0.]
# Gather indices with zeros in `zeros_1`.
res_1 = list_ops.tensor_list_gather(
zeros_1, [0], element_dtype=dtypes.float32)
# Gather indices with zeros in `zeros_2`.
res_2 = list_ops.tensor_list_gather(
zeros_2, [0, 2], element_dtype=dtypes.float32)
self.assertAllEqual(self.evaluate(res_1), [0.])
self.assertAllEqual(self.evaluate(res_2), [0., 0.])
def testZerosLikeUninitialized(self):
l0 = list_ops.tensor_list_reserve([], 3, element_dtype=dtypes.float32)
l1 = list_ops.tensor_list_set_item(l0, 0, 1.) # [1., _, _]
zeros_1 = array_ops.zeros_like(l1) # [0., _, _]
l2 = list_ops.tensor_list_set_item(l1, 2, 2.) # [1., _, 2.]
zeros_2 = array_ops.zeros_like(l2) # [0., _, 0.]
# Gather indices with zeros in `zeros_1`.
res_1 = list_ops.tensor_list_gather(
zeros_1, [0], element_dtype=dtypes.float32)
# Gather indices with zeros in `zeros_2`.
res_2 = list_ops.tensor_list_gather(
zeros_2, [0, 2], element_dtype=dtypes.float32)
self.assertAllEqual(self.evaluate(res_1), [0.])
self.assertAllEqual(self.evaluate(res_2), [0., 0.])
def testPushPopSeparateLists(self):
with self.cached_session() as sess, self.test_scope():
l = list_ops.tensor_list_reserve(
element_shape=scalar_shape(),
num_elements=20,
element_dtype=dtypes.float32)
l = list_ops.tensor_list_push_back(l, constant_op.constant(1.0))
l2 = list_ops.tensor_list_push_back(l, constant_op.constant(2.0))
l3 = list_ops.tensor_list_push_back(l, constant_op.constant(3.0))
_, e11 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
l2, e21 = list_ops.tensor_list_pop_back(l2, element_dtype=dtypes.float32)
l2, e22 = list_ops.tensor_list_pop_back(l2, element_dtype=dtypes.float32)
l3, e31 = list_ops.tensor_list_pop_back(l3, element_dtype=dtypes.float32)
l3, e32 = list_ops.tensor_list_pop_back(l3, element_dtype=dtypes.float32)
result = sess.run([e11, [e21, e22], [e31, e32]])
self.assertEqual(result, [1.0, [2.0, 1.0], [3.0, 1.0]])
def testDoNotConstantFoldVariants(self):
with self.cached_session() as sess, self.test_scope():
val = array_ops.placeholder(dtype=dtypes.float32)
l = list_ops.tensor_list_reserve(
element_shape=(7, 15), num_elements=10, element_dtype=dtypes.float32)
# Note: Pushing a Placeholder will force the constant folding code
# to build a Const node with a DT_VARIANT output. This tests that XLA
# passes a cf_consider_fn which prevent folding such nodes.
l = list_ops.tensor_list_push_back(
l, array_ops.fill(value=val, dims=(7, 15)))
l = list_ops.tensor_list_push_back(
l, constant_op.constant(2.0, shape=(7, 15)))
l, e2 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
_, e1 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
self.assertAllEqual(sess.run(e2, {val: 1.0}), 2.0 * np.ones((7, 15)))
self.assertAllEqual(sess.run(e1, {val: 1.0}), 1.0 * np.ones((7, 15)))
def testPushPopSeparateLists(self):
with self.cached_session() as sess, self.test_scope():
num = array_ops.placeholder(dtypes.int32)
l = list_ops.tensor_list_reserve(
element_shape=scalar_shape(),
num_elements=num,
element_dtype=dtypes.float32)
l = list_ops.tensor_list_push_back(l, constant_op.constant(1.0))
l2 = list_ops.tensor_list_push_back(l, constant_op.constant(2.0))
l3 = list_ops.tensor_list_push_back(l, constant_op.constant(3.0))
_, e11 = list_ops.tensor_list_pop_back(l, element_dtype=dtypes.float32)
l2, e21 = list_ops.tensor_list_pop_back(l2, element_dtype=dtypes.float32)
l2, e22 = list_ops.tensor_list_pop_back(l2, element_dtype=dtypes.float32)
l3, e31 = list_ops.tensor_list_pop_back(l3, element_dtype=dtypes.float32)
l3, e32 = list_ops.tensor_list_pop_back(l3, element_dtype=dtypes.float32)
result = sess.run([e11, [e21, e22], [e31, e32]], {num: 20})
self.assertEqual(result, [1.0, [2.0, 1.0], [3.0, 1.0]])
def __init__(self,
dtype,
size=None,
dynamic_size=None,
clear_after_read=None,
tensor_array_name=None,
handle=None,
flow=None,
infer_shape=True,
element_shape=None,
colocate_with_first_write_call=True,
name=None):
"""Constructs a graph mode TensorArray.
Args:
dtype: (required) data type of the TensorArray.
size: (optional) int32 scalar `Tensor`: the size of the TensorArray.
Required if flow is not provided.
dynamic_size: (optional) Python bool: If true, writes to the TensorArray
can grow the TensorArray past its initial size. Default: False.
clear_after_read: (optional) unused. Not supported in TensorLists.
tensor_array_name: (optional) unused.
handle: (optional) Must always be None.
flow: (optional) A variant `Tensor` scalar for a TensorList.
infer_shape: (optional, default: True) If True, shape inference is
enabled. In this case, all elements must have the same shape.
element_shape: (optional, default: None) A `TensorShape` object specifying
the shape constraints of each of the elements of the TensorArray. Need
not be fully defined.
colocate_with_first_write_call: (optional). unused.
name: (optional) A name for the operation.
Raises:
ValueError: if both handle and tensor_array_name are provided.
TypeError: if handle is provided but is not a Tensor.
"""
assert handle is None
del handle
del clear_after_read
del tensor_array_name
del colocate_with_first_write_call
self._dynamic_size = dynamic_size
if (flow is not None and
(not isinstance(flow, ops.Tensor) or flow.dtype != dtypes.variant)):
raise TypeError("flow must be a variant tensor")
if flow is None and size is None:
raise ValueError("Size must be provided if flow is not provided")
if flow is not None and size is not None:
raise ValueError("Cannot provide both a flow and size "
"at the same time")
if flow is not None and element_shape is not None:
raise ValueError("Cannot provide both a flow and element_shape "
"at the same time")
self._dtype = dtypes.as_dtype(dtype).base_dtype
# Record the current static shape for the array elements. The element
# shape is defined either by `element_shape` or the shape of the tensor
# of the first write. If `infer_shape` is true, all writes checks for
# shape equality.
if element_shape is None:
self._infer_shape = infer_shape
self._element_shape = []
else:
self._infer_shape = True
self._element_shape = [tensor_shape.as_shape(element_shape)]
with ops.name_scope(name, "TensorArrayV2", [size, flow]) as scope:
if flow is None:
self._flow = list_ops.tensor_list_reserve(
element_shape=element_shape,
num_elements=size,
element_dtype=dtype,
name=scope)
else:
self._flow = flow
# For backwards compatibility.
self._colocate_with_first_write_call = None
self._colocate_with = None
def testStackWithUninitializedTensors(self):
with self.cached_session(), self.test_scope():
l = list_ops.tensor_list_reserve(
element_dtype=dtypes.float32, element_shape=[], num_elements=3)
t = list_ops.tensor_list_stack(l, element_dtype=dtypes.float32)
self.assertAllEqual(t, [0., 0., 0.])
def testMemoryviewFailsForVariant(self):
variant_t = list_ops.tensor_list_reserve(element_shape=[],
num_elements=1,
element_dtype=dtypes.float32)
with self.assertRaisesRegex(BufferError, "Cannot convert .+ variant"):
np.asarray(memoryview(variant_t))
def __init__(self,
dtype,
size=None,
dynamic_size=None,
clear_after_read=None,
tensor_array_name=None,
handle=None,
flow=None,
infer_shape=True,
element_shape=None,
colocate_with_first_write_call=True,
name=None):
"""Constructs a graph mode TensorArray.
Args:
dtype: (required) data type of the TensorArray.
size: (optional) int32 scalar `Tensor`: the size of the TensorArray.
Required if flow is not provided.
dynamic_size: (optional) Python bool: If true, writes to the TensorArray
can grow the TensorArray past its initial size. Default: False.
clear_after_read: (optional) unused. Not supported in TensorLists.
tensor_array_name: (optional) unused.
handle: (optional) Must always be None.
flow: (optional) A variant `Tensor` scalar for a TensorList.
infer_shape: (optional, default: True) If True, shape inference is
enabled. In this case, all elements must have the same shape.
element_shape: (optional, default: None) A `TensorShape` object specifying
the shape constraints of each of the elements of the TensorArray. Need
not be fully defined.
colocate_with_first_write_call: (optional). unused.
name: (optional) A name for the operation.
Raises:
ValueError: if both handle and tensor_array_name are provided.
TypeError: if handle is provided but is not a Tensor.
"""
assert handle is None
del handle
del clear_after_read
del tensor_array_name
del colocate_with_first_write_call
self._dynamic_size = dynamic_size
if (flow is not None and
(not isinstance(flow, ops.Tensor) or flow.dtype != dtypes.variant)):
raise TypeError("flow must be a variant tensor")
if flow is None and size is None:
raise ValueError("Size must be provided if flow is not provided")
if flow is not None and size is not None:
raise ValueError("Cannot provide both a flow and size "
"at the same time")
if flow is not None and element_shape is not None:
raise ValueError("Cannot provide both a flow and element_shape "
"at the same time")
self._dtype = dtype
# Record the current static shape for the array elements. The element
# shape is defined either by `element_shape` or the shape of the tensor
# of the first write. If `infer_shape` is true, all writes checks for
# shape equality.
if element_shape is None:
self._infer_shape = infer_shape
self._element_shape = []
else:
self._infer_shape = True
self._element_shape = [tensor_shape.TensorShape(element_shape)]
with ops.name_scope(name, "TensorArrayV2", [size, flow]) as scope:
if flow is None:
self._flow = list_ops.tensor_list_reserve(
element_shape=element_shape,
num_elements=size,
element_dtype=dtype,
name=scope)
else:
self._flow = flow
# For backwards compatibility.
self._colocate_with_first_write_call = None
self._colocate_with = None
