def testAssignNoShapeNoValidateShape(self):
with self.test_session():
value = self._NewShapelessTensor()
var = state_ops.variable_op([1, 2], tf.float32, set_shape=False)
self.assertEqual(tensor_shape.unknown_shape(), var.get_shape())
self.assertEqual(tensor_shape.unknown_shape(),
tf.assign(var, value, validate_shape=False).get_shape())
def _ReductionShape(op):
"""Common shape function for reduction ops."""
input_shape = op.inputs[0].get_shape()
reduction_indices = tensor_util.constant_value(op.inputs[1])
keep_dims = op.get_attr("keep_dims")
if reduction_indices is None or input_shape.ndims is None:
if keep_dims:
return [tensor_shape.unknown_shape(ndims=input_shape.ndims)]
else:
return [tensor_shape.unknown_shape()]
# Turn reduction_indices from scalar to vector if necessary
reduction_indices = np.ravel(reduction_indices)
for reduction_index in reduction_indices:
if reduction_index < 0 or reduction_index >= input_shape.ndims:
raise ValueError("Invalid reduction dimension %d for input with %d "
"dimensions" % (reduction_index, input_shape.ndims))
returned_dims = []
if keep_dims:
for i, dim in enumerate(input_shape.dims):
if i in reduction_indices:
returned_dims.append(1)
else:
returned_dims.append(dim)
else:
for i, dim in enumerate(input_shape.dims):
if i not in reduction_indices:
returned_dims.append(dim)
return [tensor_shape.TensorShape(returned_dims)]
def testAssignNoShape(self):
with self.cached_session():
value = self._NewShapelessTensor()
var = state_ops.variable_op([1, 2], dtypes.float32, set_shape=False)
self.assertEqual(tensor_shape.unknown_shape(), var.get_shape())
self.assertEqual(tensor_shape.unknown_shape(),
state_ops.assign(var, value).get_shape())
def testPartialShapes(self):
np.random.seed(1618)
# Input shape is unknown.
reduction_axes = [1, 2]
c_unknown = tf.placeholder(tf.float32)
s_unknown = tf.reduce_sum(c_unknown, reduction_axes)
self.assertEqual(tensor_shape.unknown_shape(), s_unknown.get_shape())
np_input = np.random.randn(3, 3, 3)
self._compareAll(np_input, reduction_axes, {c_unknown: np_input})
# Input shape only has known rank.
c_known_rank = tf.placeholder(tf.float32)
c_known_rank.set_shape(tensor_shape.unknown_shape(ndims=3))
s_known_rank = tf.reduce_sum(c_known_rank, reduction_axes, keep_dims=True)
self.assertEqual(3, s_known_rank.get_shape().ndims)
np_input = np.random.randn(3, 3, 3)
self._compareAll(np_input, reduction_axes, {c_known_rank: np_input})
# Reduction indices are unknown.
unknown_indices = tf.placeholder(tf.int32)
c_unknown_indices = tf.constant([[10.0], [20.0]])
s_unknown_indices = tf.reduce_sum(c_unknown_indices, unknown_indices,
keep_dims=False)
self.assertEqual(tensor_shape.unknown_shape(),
s_unknown_indices.get_shape())
s_unknown_indices_keep = tf.reduce_sum(c_unknown_indices, unknown_indices,
keep_dims=True)
self.assertEqual(2, s_unknown_indices_keep.get_shape().ndims)
def testEquality(self):
s1 = tensor_shape.TensorShape([tensor_shape.Dimension(
3), tensor_shape.Dimension(4), tensor_shape.Dimension(7)])
s2 = tensor_shape.TensorShape([tensor_shape.Dimension(
3), tensor_shape.Dimension(4), tensor_shape.Dimension(7)])
s3 = tensor_shape.TensorShape([tensor_shape.Dimension(3),
tensor_shape.Dimension(4), None])
self.assertTrue(s1 == s2)
self.assertFalse(s1 != s2)
self.assertFalse(s1 == "a string")
self.assertTrue(s1 != "a string")
self.assertNotEqual(s1, "347", "Should not equal an ambiguous string.")
self.assertEqual(s1, ["3", "4", "7"])
# Test with an unknown shape in s3
self.assertTrue(s1 != s3)
self.assertFalse(s3 == "a string")
self.assertTrue(s3 != "a string")
# eq and neq are not symmetric for unknown shapes.
unk0 = tensor_shape.unknown_shape()
self.assertFalse(unk0 == s1)
self.assertFalse(s1 == unk0)
with self.assertRaises(ValueError):
unk0 != s1 # pylint: disable=pointless-statement
with self.assertRaises(ValueError):
s1 != unk0 # pylint: disable=pointless-statement
unk1 = tensor_shape.unknown_shape()
self.assertTrue(unk0 == unk1)
self.assertTrue(unk1 == unk0)
with self.assertRaises(ValueError):
unk0 != unk1 # pylint: disable=pointless-statement
with self.assertRaises(ValueError):
unk1 != unk0 # pylint: disable=pointless-statement
def testAsList(self):
with self.assertRaisesRegexp(ValueError,
"not defined on an unknown TensorShape"):
tensor_shape.unknown_shape().as_list()
self.assertAllEqual([None, None], tensor_shape.unknown_shape(2).as_list())
self.assertAllEqual([2, None, 4], tensor_shape.TensorShape(
(2, None, 4)).as_list())
def _SliceShape(op):
"""Shape function for array_ops.slice."""
input_shape = op.inputs[0].get_shape()
begin_shape = op.inputs[1].get_shape().with_rank_at_most(1)
sizes_shape = op.inputs[2].get_shape().with_rank_at_most(1)
rank_vector_shape = begin_shape.merge_with(sizes_shape)
ndims = rank_vector_shape.num_elements()
if ndims is not None:
input_shape.assert_has_rank(ndims)
begin_value = tensor_util.ConstantValue(op.inputs[1])
sizes_value = tensor_util.ConstantValue(op.inputs[2])
if sizes_value is not None:
returned_dims = []
for i, slice_size in enumerate(sizes_value.ravel()):
if slice_size != -1:
returned_dims.append(slice_size)
elif begin_value is not None:
returned_dims.append(input_shape[i] - begin_value[i])
else:
returned_dims.append(None)
return [tensor_shape.TensorShape(returned_dims)]
else:
if input_shape.ndims is not None:
return [tensor_shape.unknown_shape(ndims=input_shape.ndims)]
elif ndims is not None:
return [tensor_shape.unknown_shape(ndims=ndims)]
else:
return [tensor_shape.unknown_shape()]
def testAsProto(self):
self.assertTrue(tensor_shape.unknown_shape().as_proto().unknown_rank)
self.assertFalse(
tensor_shape.unknown_shape(rank=3).as_proto().unknown_rank)
self.assertFalse(
tensor_shape.TensorShape([1, 2, 3]).as_proto().unknown_rank)
self.assertFalse(
tensor_shape.TensorShape([1, None, 3]).as_proto().unknown_rank)
def _sparse_shape(op):
"""Shape function for `SparseTensor` result."""
num_rows = (op.inputs[0].get_shape()[0] if
op.type in ("DenseToSparseOperation", "DenseToDenseOperation")
else None)
return [
tensor_shape.TensorShape([num_rows, 2]),
tensor_shape.unknown_shape(1),
tensor_shape.unknown_shape(1),
]
def testStr(self):
self.assertEqual("<unknown>", str(tensor_shape.unknown_shape()))
self.assertEqual("(?,)", str(tensor_shape.unknown_shape(ndims=1)))
self.assertEqual("(?, ?)", str(tensor_shape.unknown_shape(ndims=2)))
self.assertEqual("(?, ?, ?)", str(tensor_shape.unknown_shape(ndims=3)))
self.assertEqual("()", str(tensor_shape.scalar()))
self.assertEqual("(7,)", str(tensor_shape.vector(7)))
self.assertEqual("(3, 8)", str(tensor_shape.matrix(3, 8)))
self.assertEqual("(4, 5, 2)", str(tensor_shape.TensorShape([4, 5, 2])))
self.assertEqual("(32, ?, 1, 9)",
str(tensor_shape.TensorShape([32, None, 1, 9])))
def test_build_raw_serving_input_receiver_fn_without_shape(self):
"""Test case for issue #21178."""
f = {"feature_1": array_ops.placeholder(dtypes.float32),
"feature_2": array_ops.placeholder(dtypes.int32)}
serving_input_receiver_fn = export.build_raw_serving_input_receiver_fn(f)
v = serving_input_receiver_fn()
self.assertTrue(isinstance(v, export.ServingInputReceiver))
self.assertEqual(
tensor_shape.unknown_shape(),
v.receiver_tensors["feature_1"].shape)
self.assertEqual(
tensor_shape.unknown_shape(),
v.receiver_tensors["feature_2"].shape)
def test_to_feature_columns_and_input_fn(self):
df = setup_test_df_3layer()
feature_columns, input_fn = (
estimator_utils.to_feature_columns_and_input_fn(
df,
base_input_keys_with_defaults={"a": 1,
"b": 2,
"c": 3,
"d": 4},
label_keys=["g"],
feature_keys=["a", "b", "f"]))
expected_feature_column_a = feature_column.DataFrameColumn(
"a",
learn.PredefinedSeries(
"a",
parsing_ops.FixedLenFeature(tensor_shape.unknown_shape(),
dtypes.int32, 1)))
expected_feature_column_b = feature_column.DataFrameColumn(
"b",
learn.PredefinedSeries("b", parsing_ops.VarLenFeature(dtypes.int32)))
expected_feature_column_f = feature_column.DataFrameColumn(
"f",
learn.TransformedSeries([
learn.PredefinedSeries("c",
parsing_ops.FixedLenFeature(
tensor_shape.unknown_shape(),
dtypes.int32, 3)),
learn.PredefinedSeries("d", parsing_ops.VarLenFeature(dtypes.int32))
], mocks.Mock2x2Transform("iue", "eui", "snt"), "out2"))
expected_feature_columns = [
expected_feature_column_a, expected_feature_column_b,
expected_feature_column_f
]
self.assertEqual(sorted(expected_feature_columns), sorted(feature_columns))
base_features, labels = input_fn()
expected_base_features = {
"a": mocks.MockTensor("Tensor a", dtypes.int32),
"b": mocks.MockSparseTensor("SparseTensor b", dtypes.int32),
"c": mocks.MockTensor("Tensor c", dtypes.int32),
"d": mocks.MockSparseTensor("SparseTensor d", dtypes.int32)
}
self.assertEqual(expected_base_features, base_features)
expected_labels = mocks.MockTensor("Out iue", dtypes.int32)
self.assertEqual(expected_labels, labels)
self.assertEqual(3, len(feature_columns))
def _SqueezeShape(op):
"""Determine shape for squeeze op's output tensor.
Args:
op: Operation for which to determine shape.
Returns:
Shape of op's output tensor.
Raises:
ValueError: if squeeze_dims includes a dimension outside of [-rank, rank),
where rank is the number of dimensions in the input tensor. Or, if
squeeze_dims includes a dimension for which input shape has a value
not equal to 1.
"""
input_shape = op.inputs[0].get_shape()
if input_shape.dims is None:
return [tensor_shape.unknown_shape()]
squeeze_dims = op.get_attr("squeeze_dims") or []
wrapped_squeeze_dims = []
input_ndims = input_shape.ndims
for i, squeeze_dim in enumerate(squeeze_dims):
if squeeze_dim < -input_ndims or squeeze_dim >= input_ndims:
raise ValueError(
"squeeze_dims[%d]=%d not in [%d, %d)." % (
i, squeeze_dim, -input_ndims, input_ndims))
if squeeze_dim < 0:
squeeze_dim += input_ndims
wrapped_squeeze_dims.append(squeeze_dim)
result_shape = []
for i, dim in enumerate([d.value for d in input_shape.dims]):
is_explicit_match = i in wrapped_squeeze_dims
if dim is None:
if is_explicit_match:
# Assume that the squeezed dimension will be 1 at runtime.
continue
if not wrapped_squeeze_dims:
# If squeezing all 1 dimensions and we see a None, give up.
return [tensor_shape.unknown_shape()]
elif dim == 1:
if is_explicit_match or not wrapped_squeeze_dims:
continue
elif is_explicit_match:
raise ValueError(
"Can not squeeze dim[%d], expected a dimension of 1, got %d." % (
i, dim))
result_shape.append(dim)
return [tensor_shape.TensorShape(result_shape)]
def _dense_to_dense_shape(op):
"""Shapes for `SparseTensor` result given 2 dense inputs.
Args:
op: Operation with 2 dense `Tensor` inputs.
Returns:
Tuple of three shapes corresponding to the indices, values, and shape
`Tensor` components of the result `SparseTensor`.
Raises:
ValueError: if either input `Tensor` has rank < 2, or ranks do not match, or
first n-1 dims of input shapes are not compatible.
"""
# The following should stay in sync with `ComputeDenseToDense` shape
# assertions in kernels/set_kernels.cc.
input0_shape = op.inputs[0].get_shape()
input0_rank = input0_shape.ndims
if (input0_rank is not None) and (input0_rank < 2):
raise ValueError("Input 0, expected rank >= 2, got shape %s." %
input0_shape)
# Dimension n contains the set values to be compared, so ranks and the first
# n-1 dimensions of inputs and output must match.
input1_shape = op.inputs[1].get_shape()
input1_rank = input1_shape.ndims
if (input0_rank is not None) and (input1_rank is not None) and (
input0_rank != input1_rank):
raise ValueError(
"Ranks do not match: input 0 with shape %s, input 1 with shape %s." %
(input0_shape, input1_shape))
output_rank = input1_rank if input0_rank is None else input0_rank
output_dim0 = input1_shape[1] if input0_shape[0] is None else input0_shape[0]
input0_dims = input0_shape.dims
if input0_dims is None:
group0_shape = tensor_shape.unknown_shape()
else:
group0_shape = tensor_shape.TensorShape(input0_dims[:-1])
input1_dims = input1_shape.dims
if input1_dims is None:
group1_shape = tensor_shape.unknown_shape()
else:
group1_shape = tensor_shape.TensorShape(input1_dims[:-1])
group0_shape.assert_is_compatible_with(group1_shape)
indices_shape = tensor_shape.TensorShape((output_dim0, output_rank))
values_shape = tensor_shape.unknown_shape(1)
shape_shape = tensor_shape.TensorShape((output_rank,))
return (indices_shape, values_shape, shape_shape)
def _ExpandDimsShape(op):
"""Determine shape for expand op's output tensor.
Args:
op: Operation for which to determine shape.
op.inputs[0] is the input tensor.
op.inputs[1] is the dimension in which to expand.
Returns:
Shape of op's output tensor.
Raises:
ValueError: If dim is outside of [-rank - 1, rank], where rank is the number
of dimensions in the input tensor.
"""
input_shape = op.inputs[0].get_shape()
if input_shape.dims is None:
return [tensor_shape.unknown_shape()]
dim = tensor_util.ConstantValue(op.inputs[1])
input_ndims = input_shape.ndims
if dim < -input_ndims - 1 or dim > input_ndims:
raise ValueError(
"dim %d not in [%d, %d]." % (dim, -input_ndims, input_ndims))
if dim < 0:
dim += (input_ndims + 1)
result_shape = list(input_shape.dims)
result_shape.insert(dim, 1)
return [tensor_shape.TensorShape(result_shape)]
def _DepthwiseConv2dNativeBackpropInputShape(op):
"""Shape function for the DepthwiseConv2dNativeBackpropInput op."""
input_shape = tensor_util.constant_value(op.inputs[0])
if input_shape is not None:
return [tensor_shape.TensorShape(input_shape.tolist())]
else:
return [tensor_shape.unknown_shape(ndims=4)]
def _TileShape(op):
"""Shape function for the Tile op.
This op has two inputs:
* input: A rank-N tensor.
* multiples: A length-N vector, in which the i^th element contains
the factor by which `input` will be tiled in the i^th dimension.
It has one output, which has the same rank as input, and additional
elements according to the values in multiples
Args:
op: A Tile Operation.
Returns:
A single-element list containing the shape of the output.
"""
multiples_shape = op.inputs[1].get_shape().with_rank_at_most(1)
input_shape = op.inputs[0].get_shape().with_rank(multiples_shape.num_elements())
multiples = tensor_util.ConstantValue(op.inputs[1])
if multiples is None:
return [tensor_shape.unknown_shape(ndims=input_shape.ndims)]
else:
output_dims = []
multiples = multiples.ravel()
for i, dim in enumerate(input_shape.dims):
output_dims.append(dim * multiples[i])
return [tensor_shape.TensorShape(output_dims)]
def testShape(self):
op = ops.Operation(ops._NodeDef("noop", "myop"), ops.Graph(),
[], [dtypes.float32])
t = op.outputs[0]
self.assertEqual(tensor_shape.unknown_shape(), t.get_shape())
t.set_shape([1, 2, 3])
self.assertEqual([1, 2, 3], t.get_shape())
def _AvgPool3DGradShape(op):
"""Shape function for the AvgPool3DGrad op."""
orig_input_shape = tensor_util.constant_value(op.inputs[0])
if orig_input_shape != None: # pylint:disable=g-equals-none
return [tensor_shape.TensorShape(orig_input_shape.tolist())]
else:
return [tensor_shape.unknown_shape(ndims=5)]
def _RandomShape(op):
shape_val = tensor_util.constant_value(op.inputs[0])
if shape_val is not None:
return [tensor_shape.TensorShape(shape_val)]
else:
shape_shape = op.inputs[0].get_shape().with_rank(1)
return [tensor_shape.unknown_shape(ndims=shape_shape[0].value)]
def _TransposeShape(op):
"""Shape function for the Transpose op.
This op takes two inputs:
* input: a rank-N tensor of arbitrary shape.
* shuffle: a length-N vector.
Its output is the rank-N tensor computed by permuting the dimensions
of input according to shuffle.
Args:
op: A Transpose op.
Returns:
A single-element list containing the shape of the output.
Raises:
ValueError: If the shapes of input and shuffle are incompatible.
IndexError: If shuffle contains an index that is >= the rank of input.
"""
input_shape = op.inputs[0].get_shape()
transpose_shape = op.inputs[1].get_shape().merge_with(tensor_shape.vector(
input_shape.ndims))
transpose_vec = tensor_util.ConstantValue(op.inputs[1])
if transpose_vec is None:
return [tensor_shape.unknown_shape(ndims=transpose_shape[0].value)]
else:
return [tensor_shape.TensorShape([input_shape[i]
for i in transpose_vec.tolist()])]
def constant_value_as_shape(tensor): # pylint: disable=invalid-name
"""A version of `constant_value()` that returns a `TensorShape`.
This version should be used when a constant tensor value is
interpreted as a (possibly partial) shape, e.g. in the shape
function for `tf.reshape()`. By explicitly requesting a
`TensorShape` as the return value, it is possible to represent
unknown dimensions; by contrast, `constant_value()` is
all-or-nothing.
Args:
tensor: The rank-1 Tensor to be evaluated.
Returns:
A `TensorShape` based on the constant value of the given `tensor`.
"""
shape = tensor.get_shape().with_rank(1)
if tensor.get_shape() == [0]:
return tensor_shape.scalar()
elif tensor.op.type == "Shape":
return tensor.op.inputs[0].get_shape()
elif tensor.op.type == "Pack":
ret = tensor_shape.scalar() # Empty list.
for pack_input in tensor.op.inputs:
# `pack_input` must be a scalar. Attempt to evaluate it, and append it
# to `ret`.
pack_input_val = constant_value(pack_input)
if pack_input_val is None or pack_input_val < 0:
new_dim = tensor_shape.Dimension(None)
else:
new_dim = tensor_shape.Dimension(pack_input_val)
ret = ret.concatenate([new_dim])
return ret
elif tensor.op.type == "Concat":
# We assume that `tensor.op.inputs[0]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.scalar() # Empty list.
for concat_input in tensor.op.inputs[1:]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
elif tensor.op.type == "ConcatV2":
# We assume that `tensor.op.inputs[-1]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.scalar() # Empty list.
for concat_input in tensor.op.inputs[:-1]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
else:
ret = tensor_shape.unknown_shape(shape[0].value)
value = constant_value(tensor)
if value is not None:
ret = ret.merge_with(tensor_shape.TensorShape(
[d if d != -1 else None for d in value]))
return ret
def _ConcatShape(op):
concat_dim = tensor_util.constant_value(op.inputs[0])
if concat_dim is None:
# Return an unknown shape with the same rank as the inputs, or an
# unknown rank if no input's rank is known.
rank = None
for value in op.inputs[1:]:
if rank is not None:
value.get_shape().assert_has_rank(rank)
else:
rank = value.get_shape().ndims
if rank == 0:
raise ValueError("Can't concatenate scalars (use tf.pack instead)")
return [tensor_shape.unknown_shape(ndims=rank)]
else:
# Merge all the non-concat dims, and sum the concat dim to make an
# output shape.
concat_dim = int(concat_dim)
output_shape = op.inputs[1].get_shape()
for value in op.inputs[2:]:
value_shape = value.get_shape()
if value_shape.ndims is not None and concat_dim >= value_shape.ndims:
raise ValueError("concat_dim is out of range (values rank = %d)" %
value_shape.ndims)
before = output_shape[:concat_dim].merge_with(value_shape[:concat_dim])
at = output_shape[concat_dim] + value_shape[concat_dim]
after = output_shape[
concat_dim + 1:].merge_with(value_shape[concat_dim + 1:])
output_shape = before.concatenate(at).concatenate(after)
return [output_shape]
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, n_features)
lengths: A tensor of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
input_shape = tensor_shape.unknown_shape(ndims=input_seq[0].get_shape().ndims)
for input_ in input_seq:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(input_seq)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r in result:
r.set_shape(input_shape)
return result
def testSkipEagerBuildElementShape(self):
fn = list_ops._build_element_shape
# Unknown shape -> -1.
self.assertEqual(fn(None), -1)
self.assertEqual(fn(tensor_shape.unknown_shape()), -1)
# Scalar shape -> [] with type int32.
self.assertEqual(fn([]).dtype, dtypes.int32)
self.assertEqual(fn(tensor_shape.scalar()).dtype, dtypes.int32)
self.assertAllEqual(self.evaluate(fn([])), np.array([], np.int32))
self.assertAllEqual(
self.evaluate(fn(tensor_shape.scalar())), np.array([], np.int32))
# Tensor -> Tensor
shape = constant_op.constant(1)
self.assertIs(fn(shape), shape)
# Shape with unknown dims -> shape list with -1's.
shape = [None, 5]
self.assertAllEqual(fn(shape), [-1, 5])
self.assertAllEqual(fn(tensor_shape.TensorShape(shape)), [-1, 5])
# Shape with unknown dims and tensor dims -> shape list with -1's and tensor
# dims.
t = array_ops.placeholder(dtypes.int32)
shape = [None, 5, t]
result = fn(shape)
self.assertAllEqual(result[:2], [-1, 5])
self.assertIs(result[2], t)
def _RandomShape(op):
shape_val = tensor_util.ConstantValue(op.inputs[0])
if shape_val is not None:
return [tensor_shape.TensorShape(shape_val.tolist())]
else:
shape_shape = op.inputs[0].get_shape().with_rank_at_most(1)
return [tensor_shape.unknown_shape(ndims=shape_shape.num_elements())]
def _TensorArrayReadShape(op):
# handle, index, flow_in
op.inputs[0].get_shape().merge_with(tensor_shape.vector(2))
op.inputs[1].get_shape().merge_with(tensor_shape.scalar())
op.inputs[2].get_shape().merge_with(tensor_shape.scalar())
# value
return [tensor_shape.unknown_shape()]
def testAssignNoValueShapeNoValidateShape(self): value = self._NewShapelessTensor() shape = [1, 2] var = state_ops.variable_op(shape, dtypes.float32) self.assertEqual(shape, var.get_shape()) assigned = state_ops.assign(var, value, validate_shape=False) self.assertEqual(tensor_shape.unknown_shape(), assigned.get_shape())
def _testStackWhileSwap(self, use_gpu):
with self.test_session(use_gpu=use_gpu):
n = constant_op.constant(0)
h = gen_data_flow_ops._stack(dtypes.float32, stack_name="foo")
def c(x):
return math_ops.less(x, 10)
def b(x):
with ops.control_dependencies([x]):
a = constant_op.constant(np.ones(2000), dtype=dtypes.float32)
v = gen_data_flow_ops._stack_push(h, a, swap_memory=True)
with ops.control_dependencies([v]):
return math_ops.add(x, 1)
r = control_flow_ops.while_loop(c, b, [n])
v = constant_op.constant(np.zeros(2000), dtype=dtypes.float32)
def c1(x, y):
return math_ops.greater(x, 0)
def b1(x, y):
nx = math_ops.subtract(x, 1)
ny = y + gen_data_flow_ops._stack_pop(h, dtypes.float32)
return [nx, ny]
rx, ry = control_flow_ops.while_loop(
c1, b1, [r, v], [r.get_shape(), tensor_shape.unknown_shape()])
self.assertAllClose(np.ones(2000) * 10.0, ry.eval())
def variable_op(shape, dtype, name="Variable", set_shape=True, container="",
shared_name=""):
"""Create a variable Operation.
See also variables.Variable.
Args:
shape: The shape of the tensor managed by this variable
dtype: The underlying type of the tensor values.
name: optional name to use for the variable op.
set_shape: If True, set the shape property of the returned Tensor to
the shape argument.
container: An optional string. Defaults to "".
If non-empty, this variable is placed in the given container.
Otherwise, a default container is used.
shared_name: An optional string. Defaults to "".
If non-empty, this variable is named in the given bucket
with this shared_name. Otherwise, the node name is used instead.
Returns:
A variable tensor.
"""
if not set_shape:
shape = tensor_shape.unknown_shape()
ret = gen_state_ops._variable(shape=shape, dtype=dtype, name=name,
container=container, shared_name=shared_name)
# TODO(mrry): Move this to where it is used, so we can get rid of this op
# wrapper?
if set_shape:
ret.set_shape(shape)
return ret
def accumulate_n(inputs, shape=None, tensor_dtype=None, name=None):
"""Returns the element-wise sum of a list of tensors.
Optionally, pass `shape` and `tensor_dtype` for shape and type checking,
otherwise, these are inferred.
For example:
```python
# tensor 'a' is [[1, 2], [3, 4]
# tensor `b` is [[5, 0], [0, 6]]
tf.accumulate_n([a, b, a]) ==> [[7, 4], [6, 14]]
# Explicitly pass shape and type
tf.accumulate_n([a, b, a], shape=[2, 2], tensor_dtype=tf.int32)
==> [[7, 4], [6, 14]]
```
Args:
inputs: A list of `Tensor` objects, each with same shape and type.
shape: Shape of elements of `inputs`.
tensor_dtype: The type of `inputs`.
name: A name for the operation (optional).
Returns:
A `Tensor` of same shape and type as the elements of `inputs`.
Raises:
ValueError: If `inputs` don't all have same shape and dtype or the shape
cannot be inferred.
"""
if tensor_dtype is None:
if not inputs or not isinstance(inputs, (list, tuple)):
raise ValueError(
"inputs must be a list of at least one Tensor with the "
"same dtype and shape")
inputs = ops.convert_n_to_tensor_or_indexed_slices(inputs)
if not all(isinstance(x, ops.Tensor) for x in inputs):
raise ValueError(
"inputs must be a list of at least one Tensor with the "
"same dtype and shape")
if not all(x.dtype == inputs[0].dtype for x in inputs):
raise ValueError(
"inputs must be a list of at least one Tensor with the "
"same dtype and shape")
tensor_dtype = inputs[0].dtype
if shape is not None:
shape = tensor_shape.as_shape(shape)
else:
shape = tensor_shape.unknown_shape()
for input_tensor in inputs:
if isinstance(input_tensor, ops.Tensor):
shape = shape.merge_with(input_tensor.get_shape())
if not shape.is_fully_defined():
# TODO(pbar): Make a version of assign_add that accepts an uninitialized
# lvalue, and takes its shape from that? This would allow accumulate_n to
# work in all situations that add_n currently works.
raise ValueError(
"Cannot infer the shape of the accumulator for "
"accumulate_n. Pass the shape argument, or set the shape "
"of at least one of the inputs.")
with ops.op_scope(inputs, name, "AccumulateN") as name:
var = gen_state_ops._temporary_variable(shape=shape,
dtype=tensor_dtype)
var_name = var.op.name
var = state_ops.assign(var, array_ops.zeros_like(inputs[0]))
update_ops = []
for input_tensor in inputs:
op = state_ops.assign_add(var, input_tensor, use_locking=True)
update_ops.append(op)
with ops.control_dependencies(update_ops):
return gen_state_ops._destroy_temporary_variable(var,
var_name=var_name,
name=name)
def constant_value_as_shape(tensor): # pylint: disable=invalid-name
"""A version of `constant_value()` that returns a `TensorShape`.
This version should be used when a constant tensor value is
interpreted as a (possibly partial) shape, e.g. in the shape
function for `tf.reshape()`. By explicitly requesting a
`TensorShape` as the return value, it is possible to represent
unknown dimensions; by contrast, `constant_value()` is
all-or-nothing.
Args:
tensor: The rank-1 Tensor to be evaluated.
Returns:
A `TensorShape` based on the constant value of the given `tensor`.
"""
if context.in_eager_mode():
return tensor_shape.as_shape(
[dim if dim != -1 else None for dim in tensor.numpy()])
shape = tensor.get_shape().with_rank(1)
if tensor.get_shape() == [0]:
return tensor_shape.scalar()
elif tensor.op.type == "Shape":
return tensor.op.inputs[0].get_shape()
elif tensor.op.type == "Pack":
ret = tensor_shape.scalar() # Empty list.
# Since we expect rank 1 inputs, Pack's axis must be zero, otherwise it
# would not be rank 1.
assert tensor.op.get_attr("axis") == 0
for pack_input in tensor.op.inputs:
# `pack_input` must be a scalar. Attempt to evaluate it, and append it
# to `ret`.
pack_input_val = constant_value(pack_input)
if pack_input_val is None or pack_input_val < 0:
new_dim = tensor_shape.Dimension(None)
else:
new_dim = tensor_shape.Dimension(pack_input_val)
ret = ret.concatenate([new_dim])
return ret
elif tensor.op.type == "Concat":
# We assume that `tensor.op.inputs[0]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.scalar() # Empty list.
for concat_input in tensor.op.inputs[1:]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
elif tensor.op.type == "ConcatV2":
# We assume that `tensor.op.inputs[-1]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.scalar() # Empty list.
for concat_input in tensor.op.inputs[:-1]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
elif tensor.op.type == "StridedSlice":
try:
begin = constant_value(tensor.op.inputs[1])
end = constant_value(tensor.op.inputs[2])
strides = constant_value(tensor.op.inputs[3])
if begin is not None and end is not None and strides is not None:
begin = begin[0]
end = end[0]
strides = strides[0]
begin_mask = tensor.op.get_attr("begin_mask")
if begin_mask == 1:
begin = None
end_mask = tensor.op.get_attr("end_mask")
if end_mask == 1:
end = None
ellipsis_mask = tensor.op.get_attr("ellipsis_mask")
new_axis_mask = tensor.op.get_attr("new_axis_mask")
shrink_axis_mask = tensor.op.get_attr("shrink_axis_mask")
valid_attributes = (not ellipsis_mask and not new_axis_mask
and not shrink_axis_mask
and (not begin_mask or (begin_mask == 1))
and (not end_mask or (end_mask == 1)))
if valid_attributes: # additional inputs not supported
prev = constant_value_as_shape(tensor.op.inputs[0])
prev = prev[begin:end:strides]
ret = tensor_shape.TensorShape(prev)
return ret
except ValueError: # Could come from get_attr or slicing prev.
pass
except TypeError: # Could come from slicing prev.
pass
ret = tensor_shape.unknown_shape(shape[0].value)
value = constant_value(tensor)
if value is not None:
ret = ret.merge_with(
tensor_shape.TensorShape([d if d >= 0 else None for d in value]))
return ret
def _TensorArrayConcatShape(op):
# handle, flow_in
op.inputs[0].get_shape().merge_with(tensor_shape.vector(2))
op.inputs[1].get_shape().merge_with(tensor_shape.scalar())
# value, lengths
return [tensor_shape.unknown_shape(), tensor_shape.vector(None)]
def _NewShapelessTensor(self):
tensor = tf.placeholder(tf.float32)
self.assertEqual(tensor_shape.unknown_shape(), tensor.get_shape())
return tensor
def testAssignUpdateNoShape(self): var = state_ops.variable_op([1, 2], tf.float32, set_shape=False) added = tf.assign_add(var, self._NewShapelessTensor()) self.assertEqual(tensor_shape.unknown_shape(), added.get_shape()) subbed = tf.assign_sub(var, self._NewShapelessTensor()) self.assertEqual(tensor_shape.unknown_shape(), subbed.get_shape())
def _AccumulatorShape(inputs):
shape = tensor_shape.unknown_shape()
for i in inputs:
if isinstance(i, ops.Tensor):
shape = shape.merge_with(i.get_shape())
return shape
def _LookupTableExportShape(op):
"""Shape function for data_flow_ops._lookup_table_export_values."""
op.inputs[0].get_shape().merge_with(tensor_shape.scalar())
keys_shape = tensor_shape.vector(None)
values_shape = tensor_shape.unknown_shape()
return [keys_shape, values_shape]
def _MatchingFilesShape(op):
"""Shape function for the MatchingFiles op."""
unused_patern_shape = op.inputs[0].get_shape().merge_with(
tensor_shape.scalar())
return [tensor_shape.unknown_shape(ndims=1)]
def _ReaderReadUpToShape(_):
"""Shape function for the ReaderBase.ReadUpTo op."""
return [
tensor_shape.unknown_shape(ndims=1),
tensor_shape.unknown_shape(ndims=1)
]
def unknown_shape(op):
"""Shape function for use with ops whose output shapes are unknown."""
return [tensor_shape.unknown_shape() for _ in op.outputs]
def f():
c = lambda n: n < 10
b = lambda n: n * x
return control_flow_ops.while_loop(
c, b, [n], [tensor_shape.unknown_shape()])
def testset_shape(self):
p = state_ops.variable_op([1, 2], dtypes.float32)
self.assertEqual([1, 2], p.get_shape())
p = state_ops.variable_op([1, 2], dtypes.float32, set_shape=False)
self.assertEqual(tensor_shape.unknown_shape(), p.get_shape())
def testAssignNoVarShapeNoValidateShape(self):
value = np.array([[42.0, 43.0]])
var = state_ops.variable_op(value.shape, tf.float32, set_shape=False)
self.assertEqual(tensor_shape.unknown_shape(), var.get_shape())
assigned = tf.assign(var, value, validate_shape=False)
self.assertShapeEqual(value, assigned)
def _TensorArrayPackShape(op):
# handle, flow_in
op.inputs[0].get_shape().merge_with(tensor_shape.vector(2))
op.inputs[1].get_shape().merge_with(tensor_shape.scalar())
# value
return [tensor_shape.unknown_shape()]
def testAsDenseShapes(self):
test_cases = (
{
"types": (),
"classes": (),
"expected": ()
},
{
"types": tensor_shape.scalar(),
"classes": ops.Tensor,
"expected": tensor_shape.scalar()
},
{
"types": tensor_shape.scalar(),
"classes": sparse_tensor.SparseTensor,
"expected": tensor_shape.unknown_shape()
},
{
"types": (tensor_shape.scalar()),
"classes": (ops.Tensor),
"expected": (tensor_shape.scalar())
},
{
"types": (tensor_shape.scalar()),
"classes": (sparse_tensor.SparseTensor),
"expected": (tensor_shape.unknown_shape())
},
{
"types": (tensor_shape.scalar(), ()),
"classes": (ops.Tensor, ()),
"expected": (tensor_shape.scalar(), ())
},
{
"types": ((), tensor_shape.scalar()),
"classes": ((), ops.Tensor),
"expected": ((), tensor_shape.scalar())
},
{
"types": (tensor_shape.scalar(), ()),
"classes": (sparse_tensor.SparseTensor, ()),
"expected": (tensor_shape.unknown_shape(), ())
},
{
"types": ((), tensor_shape.scalar()),
"classes": ((), sparse_tensor.SparseTensor),
"expected": ((), tensor_shape.unknown_shape())
},
{
"types": (tensor_shape.scalar(), (), tensor_shape.scalar()),
"classes": (ops.Tensor, (), ops.Tensor),
"expected": (tensor_shape.scalar(), (), tensor_shape.scalar())
},
{
"types": (tensor_shape.scalar(), (), tensor_shape.scalar()),
"classes": (sparse_tensor.SparseTensor, (),
sparse_tensor.SparseTensor),
"expected": (tensor_shape.unknown_shape(), (),
tensor_shape.unknown_shape())
},
{
"types": ((), tensor_shape.scalar(), ()),
"classes": ((), ops.Tensor, ()),
"expected": ((), tensor_shape.scalar(), ())
},
{
"types": ((), tensor_shape.scalar(), ()),
"classes": ((), sparse_tensor.SparseTensor, ()),
"expected": ((), tensor_shape.unknown_shape(), ())
},
)
for test_case in test_cases:
self.assertShapesEqual(
sparse.as_dense_shapes(test_case["types"], test_case["classes"]),
test_case["expected"])
def _NewShapelessTensor(self):
tensor = array_ops.placeholder(dtypes.float32)
self.assertEqual(tensor_shape.unknown_shape(), tensor.get_shape())
return tensor
def _flat_shapes(self):
# NOTE(mrry): The default flat shape of a boxed `SparseTensor` is `(3,)`,
# but a `SparseTensorStructure` can also represent a batch of boxed
# `SparseTensor` objects with shape `(?, 3)` (and batches of batches, etc.),
# so the flat shape must be unknown.
return [tensor_shape.unknown_shape(None)]
def _LookupTableFindShape(op):
"""Shape function for data_flow_ops._lookup_table_find."""
op.inputs[0].get_shape().merge_with(tensor_shape.scalar())
return [tensor_shape.unknown_shape()]
def constant_value_as_shape(tensor): # pylint: disable=invalid-name
"""A version of `constant_value()` that returns a `TensorShape`.
This version should be used when a constant tensor value is
interpreted as a (possibly partial) shape, e.g. in the shape
function for `tf.reshape()`. By explicitly requesting a
`TensorShape` as the return value, it is possible to represent
unknown dimensions; by contrast, `constant_value()` is
all-or-nothing.
Args:
tensor: The rank-1 Tensor to be evaluated.
Returns:
A `TensorShape` based on the constant value of the given `tensor`.
"""
shape = tensor.get_shape().with_rank(1)
if tensor.get_shape() == [0]:
return tensor_shape.scalar()
elif tensor.op.type == "Shape":
return tensor.op.inputs[0].get_shape()
elif tensor.op.type == "Pack":
ret = tensor_shape.scalar() # Empty list.
for pack_input in tensor.op.inputs:
# `pack_input` must be a scalar. Attempt to evaluate it, and append it
# to `ret`.
pack_input_val = constant_value(pack_input)
if pack_input_val is None or pack_input_val < 0:
new_dim = tensor_shape.Dimension(None)
else:
new_dim = tensor_shape.Dimension(pack_input_val)
ret = ret.concatenate([new_dim])
return ret
elif tensor.op.type == "Concat":
# We assume that `tensor.op.inputs[0]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.scalar() # Empty list.
for concat_input in tensor.op.inputs[1:]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
elif tensor.op.type == "ConcatV2":
# We assume that `tensor.op.inputs[-1]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.scalar() # Empty list.
for concat_input in tensor.op.inputs[:-1]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
else:
ret = tensor_shape.unknown_shape(shape[0].value)
value = constant_value(tensor)
if value is not None:
ret = ret.merge_with(
tensor_shape.TensorShape(
[d if d != -1 else None for d in value]))
return ret
def get_shape(self):
return tensor_shape.unknown_shape()
def _PlaceholderShape(op):
given_shape = tensor_util.TensorShapeProtoToList(op.get_attr("shape"))
if given_shape:
return [tensor_shape.TensorShape(given_shape)]
else:
return [tensor_shape.unknown_shape()]
def _AddNShape(op):
merged_shape = tensor_shape.unknown_shape()
for input_ in op.inputs:
merged_shape = merged_shape.merge_with(input_.get_shape())
return [merged_shape]
def safe_embedding_lookup_sparse(
embedding_weights,
sparse_ids,
sparse_weights=None,
combiner="mean",
default_id=None,
name="safe_embedding_lookup_sparse",
partition_strategy=None, # no used
max_norm=None,
return_trainable=False,
):
"""Provides a dynamic version of `tf.nn.safe_embedding_lookup_sparse`.
Lookup embedding results, accounting for empty features and invalid weights.
Any IDs will be treated as valid include non-positive IDs.
Invalid weights (<= 0) are pruned from input weights, as well as any IDs
with non-positive weight. For an entry with no features, the embedding vector
for `default_id` is returned, or the 0-vector if `default_id` is not supplied.
The ids and weights may be multi-dimensional. Embeddings are always aggregated
along the last dimension.
Args:
embedding_weights: A single `dynamic_embedding.Variable` instance
representing the complete embedding tensor.
sparse_ids: `SparseTensor` of shape `[d_0, d_1, ..., d_n]` containing the
ids. `d_0` is typically batch size.
sparse_weights: `SparseTensor` of same shape as `sparse_ids`, containing
float weights corresponding to `sparse_ids`, or `None` if all weights are
be assumed to be 1.0.
combiner: A string specifying how to combine embedding results for each
entry. Currently "mean", "sqrtn" and "sum" are supported, with "mean" the
default.
default_id: The id to use for an entry with no features.
name: A name for this operation (optional).
partition_strategy: A string specifying the partitioning strategy. Currently
`"div"` and `"mod"` are supported. Default is `"div"`.
max_norm: If not `None`, all embeddings are l2-normalized to max_norm before
combining.
Returns:
combined_embeddings:
A dense `Tensor` of shape `[d_0, d_1, ..., d_{n-1}, e_1, ..., e_m]`.
trainable_wrap:
A TrainableWrapper object used to fill the Optimizers `var_list`
Only provided if `return_trainable` is True.
Raises:
ValueError: if `embedding_weights` is empty.
"""
if embedding_weights is None:
raise ValueError("Missing embedding_weights %s." % embedding_weights)
if embedding_weights.key_dtype != sparse_ids.dtype:
raise TypeError(
"embedding_weights.key_dtype should be same with sparse_ids.dtype: "
"{} vs. {}".format(embedding_weights.key_dtype, sparse_ids.dtype))
weights_dtype = sparse_weights.dtype if sparse_weights is not None else None
if weights_dtype and embedding_weights.value_dtype != weights_dtype:
raise TypeError(
"embedding_weights.value_dtype should be same with sparse_weights.dtype"
": {} vs. {}".format(embedding_weights.value_dtype, weights_dtype))
scope = variable_scope.get_variable_scope()
full_name = scope.name + "/" + name if scope.name else name
with ops.name_scope(full_name + "/"):
# Reshape higher-rank sparse ids and weights to linear segment ids.
original_shape = sparse_ids.dense_shape
original_rank_dim = tensor_shape.dimension_value(
sparse_ids.dense_shape.get_shape()[0])
original_rank = (array_ops.size(original_shape)
if original_rank_dim is None else original_rank_dim)
sparse_ids = sparse_ops.sparse_reshape(
sparse_ids,
[
math_ops.reduce_prod(
array_ops.slice(original_shape, [0], [original_rank - 1])),
array_ops.gather(original_shape, original_rank - 1),
],
)
if sparse_weights is not None:
sparse_weights = sparse_tensor.SparseTensor(
sparse_ids.indices, sparse_weights.values,
sparse_ids.dense_shape)
# Prune invalid weights.
if combiner != "sum":
sparse_ids, sparse_weights = _prune_invalid_weights(
sparse_ids, sparse_weights)
# Fill in dummy values for empty features, if necessary.
sparse_ids, is_row_empty = sparse_ops.sparse_fill_empty_rows(
sparse_ids, default_id or 0)
if sparse_weights is not None:
sparse_weights, _ = sparse_ops.sparse_fill_empty_rows(
sparse_weights, 1.0)
result, trainable_ = embedding_lookup_sparse(
embedding_weights,
sparse_ids,
sparse_weights,
combiner=combiner,
partition_strategy=partition_strategy,
name=name + "/embedding_lookup_sparse",
max_norm=max_norm,
return_trainable=True,
)
if default_id is None:
# Broadcast is_row_empty to the same shape as embedding_lookup_result,
# for use in Select.
is_row_empty = array_ops.tile(
array_ops.reshape(is_row_empty, [-1, 1]),
array_ops.stack([1, array_ops.shape(result)[1]]),
)
result = array_ops.where(is_row_empty,
array_ops.zeros_like(result),
result,
name="where")
# Reshape back from linear ids back into higher-dimensional dense result.
final_result = array_ops.reshape(
result,
array_ops.concat(
[
array_ops.slice(
math_ops.cast(original_shape, dtypes.int32),
[0],
[original_rank - 1],
),
array_ops.slice(array_ops.shape(result), [1], [-1]),
],
0,
),
)
final_result.set_shape(
tensor_shape.unknown_shape(
(tensor_shape.Dimension(original_rank_dim) -
1).value).concatenate(result.get_shape()[1:]))
return (final_result, trainable_) if return_trainable else final_result
def constant_value_as_shape(tensor): # pylint: disable=invalid-name
"""A version of `constant_value()` that returns a `TensorShape`.
This version should be used when a constant tensor value is
interpreted as a (possibly partial) shape, e.g. in the shape
function for `tf.reshape()`. By explicitly requesting a
`TensorShape` as the return value, it is possible to represent
unknown dimensions; by contrast, `constant_value()` is
all-or-nothing.
Args:
tensor: The rank-0 or rank-1 Tensor to be evaluated.
Returns:
A `TensorShape` based on the constant value of the given `tensor`.
Raises:
ValueError: If the shape is rank-0 and is not statically known to be -1.
"""
if isinstance(tensor, ops.EagerTensor):
return tensor_shape.as_shape(
[dim if dim != -1 else None for dim in tensor.numpy()])
if tensor.get_shape().ndims == 0:
value = constant_value(tensor)
if value is None:
raise ValueError(
"Received a scalar with unknown value as shape; require a statically "
"known scalar with value '-1' to describe an unknown shape.")
if value != -1:
raise ValueError(
"Received a scalar value '%s' as shape; require a statically known "
"scalar with value '-1' to describe an unknown shape." % value)
return tensor_shape.unknown_shape()
shape = tensor.get_shape().with_rank(1)
if shape == [0]:
return tensor_shape.TensorShape([])
elif tensor.op.type == "Cast":
pre_cast = constant_value_as_shape(tensor.op.inputs[0])
if pre_cast.dims is None:
# the input to cast has a totally undefined shape; just return that.
return pre_cast
cast_dtype = dtypes.as_dtype(tensor.op.get_attr("DstT"))
if cast_dtype not in (dtypes.int32, dtypes.int64):
return tensor_shape.unknown_shape(shape.dims[0].value)
dest_dtype_shape_array = np.array([
x if x is not None else -1 for x in pre_cast.as_list()
]).astype(cast_dtype.as_numpy_dtype)
return tensor_shape.TensorShape(
[x if x >= 0 else None for x in dest_dtype_shape_array])
elif tensor.op.type == "Shape":
return tensor.op.inputs[0].get_shape()
elif tensor.op.type == "Pack":
ret = tensor_shape.TensorShape([]) # Empty list.
# Since we expect rank 1 inputs, Pack's axis must be zero, otherwise it
# would not be rank 1.
assert tensor.op.get_attr("axis") == 0
for pack_input in tensor.op.inputs:
# `pack_input` must be a scalar. Attempt to evaluate it, and append it
# to `ret`.
pack_input_val = constant_value(pack_input)
if pack_input_val is None or pack_input_val < 0:
new_dim = tensor_shape.Dimension(None)
else:
new_dim = tensor_shape.Dimension(pack_input_val)
ret = ret.concatenate([new_dim])
return ret
elif tensor.op.type == "Concat":
# We assume that `tensor.op.inputs[0]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.TensorShape([]) # Empty list.
for concat_input in tensor.op.inputs[1:]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
elif tensor.op.type == "ConcatV2":
# We assume that `tensor.op.inputs[-1]` evaluates to 0, as this is
# the only legal value when concatenating vectors, and it will
# have been checked by a previous shape function.
ret = tensor_shape.TensorShape([]) # Empty list.
for concat_input in tensor.op.inputs[:-1]:
# `concat_input` must be a vector. Attempt to evaluate it as a shape,
# and concatenate it with `ret`.
ret = ret.concatenate(constant_value_as_shape(concat_input))
return ret
elif tensor.op.type == "StridedSlice":
try:
begin = constant_value(tensor.op.inputs[1])
end = constant_value(tensor.op.inputs[2])
strides = constant_value(tensor.op.inputs[3])
if begin is not None and end is not None and strides is not None:
begin = begin[0]
end = end[0]
strides = strides[0]
begin_mask = tensor.op.get_attr("begin_mask")
if begin_mask == 1:
begin = None
end_mask = tensor.op.get_attr("end_mask")
if end_mask == 1:
end = None
ellipsis_mask = tensor.op.get_attr("ellipsis_mask")
new_axis_mask = tensor.op.get_attr("new_axis_mask")
shrink_axis_mask = tensor.op.get_attr("shrink_axis_mask")
valid_attributes = (not ellipsis_mask and not new_axis_mask
and not shrink_axis_mask
and (not begin_mask or (begin_mask == 1))
and (not end_mask or (end_mask == 1)))
if valid_attributes: # additional inputs not supported
prev = constant_value_as_shape(tensor.op.inputs[0])
prev = prev[begin:end:strides]
ret = tensor_shape.TensorShape(prev)
return ret
except ValueError: # Could come from get_attr or slicing prev.
pass
except TypeError: # Could come from slicing prev.
pass
elif (tensor.op.type == "Placeholder" and tensor.op.graph.building_function
and hasattr(tensor.op.graph, "internal_captures")):
# If we are inside a FuncGraph try to lookup the constant value of the
# corresponding external capture. Note that we only look at captures and
# not the fed inputs because those can be fed different values in different
# instantiations of the function call or different iterations of a
# tf.while_loop.
for i, capture in enumerate(tensor.op.graph.internal_captures):
if capture is tensor:
external_capture = tensor.op.graph.external_captures[i]
return constant_value_as_shape(external_capture)
ret = tensor_shape.unknown_shape(shape.dims[0].value)
value = constant_value(tensor)
if value is not None:
ret = ret.merge_with(
tensor_shape.TensorShape([d if d >= 0 else None for d in value]))
return ret
def __init__(self,
initial_value=None,
name=None,
trainable=True,
collections=None,
dtype=None,
shape=None):
"""Creates a variable.
Args:
initial_value: An `Output` or Python object convertible to an `Output`
representing the initial value of this variable.
name: The name of this variable. Automatically uniquified.
trainable: Whether the global read of this variable will be used for
training.
collections: Additional collections to which the `read` operation for
this variable is to be added. Defaults to [].
dtype: The type of this variable. Can be omitted if it can be deduced
from the initial_value. If different from the type of the initial
value it will be cast to this type.
shape: The shape of this variable. Only specify if there is no initial
value but shape inference is desired.
"""
if initial_value is not None:
initial_value = ops.convert_to_tensor(initial_value)
if dtype is None:
assert initial_value is not None, (
"Trying to create a resource variable "
"with no dtype or initial value. At"
" least one of these must be set.")
dtype = initial_value.dtype
elif initial_value is not None:
initial_value = math_ops.cast(initial_value, dtype)
if shape is None:
if initial_value is not None:
shape = initial_value.get_shape().as_proto()
else:
shape = tensor_shape.unknown_shape()
else:
shape = tensor_shape.as_shape(shape)
self._dtype = dtype
with ops.name_scope(name, "Variable", [initial_value]) as name:
self._handle = gen_resource_variable_ops.var_handle_op(
shared_name=name, name=name, dtype=dtype, shape=shape)
with ops.name_scope("IsInitialized"):
self._is_initialized_op = (
gen_resource_variable_ops.var_is_initialized_op(
self._handle))
if initial_value is not None:
with ops.name_scope("Create"):
self._initialize_op = gen_resource_variable_ops.create_variable_op(
self._handle, initial_value)
resources.register_resource(self._handle, self._initialize_op,
self._is_initialized_op)
with ops.name_scope("Read"):
self._value = gen_resource_variable_ops.read_variable_op(
self._handle, dtype=self._dtype)
_register_variable_read(self._value,
trainable=trainable,
collections=collections)
def get_observation_model(self, times):
parent_model = super(UnknownShapeModel,
self).get_observation_model(times)
return array_ops.placeholder_with_default(
input=parent_model, shape=tensor_shape.unknown_shape())
def safe_embedding_lookup_sparse(embedding_weights,
sparse_ids,
sparse_weights=None,
combiner='mean',
default_id=None,
name=None,
partition_strategy='div',
max_norm=None):
"""Lookup embedding results, accounting for invalid IDs and empty features.
The partitioned embedding in `embedding_weights` must all be the same shape
except for the first dimension. The first dimension is allowed to vary as the
vocabulary size is not necessarily a multiple of `P`. `embedding_weights`
may be a `PartitionedVariable` as returned by using `tf.get_variable()` with a
partitioner.
Invalid IDs (< 0) are pruned from input IDs and weights, as well as any IDs
with non-positive weight. For an entry with no features, the embedding vector
for `default_id` is returned, or the 0-vector if `default_id` is not supplied.
The ids and weights may be multi-dimensional. Embeddings are always aggregated
along the last dimension.
Args:
embedding_weights: A list of `P` float `Tensor`s or values representing
partitioned embedding `Tensor`s. Alternatively, a `PartitionedVariable`
created by partitioning along dimension 0. The total unpartitioned
shape should be `[e_0, e_1, ..., e_m]`, where `e_0` represents the
vocab size and `e_1, ..., e_m` are the embedding dimensions.
sparse_ids: `SparseTensor` of shape `[d_0, d_1, ..., d_n]` containing the
ids. `d_0` is typically batch size.
sparse_weights: `SparseTensor` of same shape as `sparse_ids`, containing
float weights corresponding to `sparse_ids`, or `None` if all weights
are be assumed to be 1.0.
combiner: A string specifying how to combine embedding results for each
entry. Currently "mean", "sqrtn" and "sum" are supported, with "mean"
the default.
default_id: The id to use for an entry with no features.
name: A name for this operation (optional).
partition_strategy: A string specifying the partitioning strategy.
Currently `"div"` and `"mod"` are supported. Default is `"div"`.
max_norm: If not `None`, all embeddings are l2-normalized to max_norm before
combining.
Returns:
Dense `Tensor` of shape `[d_0, d_1, ..., d_{n-1}, e_1, ..., e_m]`.
Raises:
ValueError: if `embedding_weights` is empty.
"""
if embedding_weights is None:
raise ValueError('Missing embedding_weights %s.' % embedding_weights)
if isinstance(embedding_weights, variables.PartitionedVariable):
embedding_weights = list(
embedding_weights) # get underlying Variables.
if not isinstance(embedding_weights, list):
embedding_weights = [embedding_weights]
if len(embedding_weights) < 1:
raise ValueError('Missing embedding_weights %s.' % embedding_weights)
dtype = sparse_weights.dtype if sparse_weights is not None else None
embedding_weights = [
ops.convert_to_tensor(w, dtype=dtype) for w in embedding_weights
]
with ops.name_scope(name, 'embedding_lookup', embedding_weights +
[sparse_ids, sparse_weights]) as scope:
# Reshape higher-rank sparse ids and weights to linear segment ids.
original_shape = sparse_ids.dense_shape
original_rank_dim = sparse_ids.dense_shape.get_shape()[0]
original_rank = (array_ops.size(original_shape)
if original_rank_dim.value is None else
original_rank_dim.value)
sparse_ids = sparse_ops.sparse_reshape(sparse_ids, [
math_ops.reduce_prod(
array_ops.slice(original_shape, [0], [original_rank - 1])),
array_ops.gather(original_shape, original_rank - 1)
])
if sparse_weights is not None:
sparse_weights = sparse_tensor.SparseTensor(
sparse_ids.indices, sparse_weights.values,
sparse_ids.dense_shape)
# Prune invalid ids and weights.
sparse_ids, sparse_weights = _prune_invalid_ids(
sparse_ids, sparse_weights)
if combiner != 'sum':
sparse_ids, sparse_weights = _prune_invalid_weights(
sparse_ids, sparse_weights)
# Fill in dummy values for empty features, if necessary.
sparse_ids, is_row_empty = sparse_ops.sparse_fill_empty_rows(
sparse_ids, default_id or 0)
if sparse_weights is not None:
sparse_weights, _ = sparse_ops.sparse_fill_empty_rows(
sparse_weights, 1.0)
result = embedding_lookup_sparse(
embedding_weights,
sparse_ids,
sparse_weights,
combiner=combiner,
partition_strategy=partition_strategy,
name=None if default_id is None else scope,
max_norm=max_norm)
if default_id is None:
# Broadcast is_row_empty to the same shape as embedding_lookup_result,
# for use in Select.
is_row_empty = array_ops.tile(
array_ops.reshape(is_row_empty, [-1, 1]),
array_ops.stack([1, array_ops.shape(result)[1]]))
result = array_ops.where(is_row_empty,
array_ops.zeros_like(result),
result,
name=scope)
# Reshape back from linear ids back into higher-dimensional dense result.
final_result = array_ops.reshape(
result,
array_ops.concat([
array_ops.slice(math_ops.cast(original_shape, dtypes.int32),
[0], [original_rank - 1]),
array_ops.slice(array_ops.shape(result), [1], [-1])
], 0))
final_result.set_shape(
tensor_shape.unknown_shape(
(original_rank_dim - 1).value).concatenate(
result.get_shape()[1:]))
return final_result
def testSlice(self):
tf_val = array_ops.placeholder(dtypes.int32, shape=(4, ))[0:2]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([None, None], c_val.as_list())
# begin:end
tf_val = constant_op.constant([10, 20, 30])[1:3]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([20, 30], c_val.as_list())
# begin:end:stride
tf_val = array_ops.strided_slice(constant_op.constant([10, 20, 30]),
[1], [3],
strides=[2])
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([20], c_val.as_list())
# [1, 2, 16, 37, None, 48]
tf_val_orig = array_ops.concat(
[[1, 2, 16, 37],
array_ops.placeholder(dtypes.int32, shape=(1, )), [48]], 0)
# begin: no end
tf_val = tf_val_orig[2:]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([16, 37, None, 48], c_val.as_list())
# begin::negative slice
tf_val = tf_val_orig[2::-1]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([16, 2, 1], c_val.as_list())
# :end:negative slice
tf_val = tf_val_orig[:1:-2]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([48, 37], c_val.as_list())
# begin:end:negative slice
tf_val = tf_val_orig[3:1:-1]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([37, 16], c_val.as_list())
# begin:negative end:slice
tf_val = tf_val_orig[1:-3:1]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([2, 16], c_val.as_list())
# negative begin::slice
tf_val = tf_val_orig[-3::1]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([37, None, 48], c_val.as_list())
# negative begin::negative slice
tf_val = tf_val_orig[-3::-1]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([37, 16, 2, 1], c_val.as_list())
# negative begin:negative end:negative slice
tf_val = tf_val_orig[-3:-5:-1]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([37, 16], c_val.as_list())
# Do not support shape inference for additional arguments
tf_val = constant_op.constant([10, 20, 30])[...]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual([None, None, None], c_val.as_list())
# Do not support shape inference for tensor slices.
tf_val = constant_op.constant(
[10, 20, 30])[array_ops.placeholder(dtypes.int32, shape=()):]
c_val = tensor_util.constant_value_as_shape(tf_val)
self.assertEqual(tensor_shape.unknown_shape(), c_val)
# Do not support shape inference for higher rank
with self.assertRaises(ValueError):
tf_val = constant_op.constant([[10], [20], [30]])[:, 0:]
c_val = tensor_util.constant_value_as_shape(tf_val)
def safe_embedding_lookup_sparse(embedding_weights,
sparse_ids,
sparse_weights=None,
combiner="mean",
default_id=None,
name=None,
partition_strategy="div",
max_norm=None):
"""Lookup embedding results, accounting for invalid IDs and empty features.
The partitioned embedding in `embedding_weights` must all be the same shape
except for the first dimension. The first dimension is allowed to vary as the
vocabulary size is not necessarily a multiple of `P`. `embedding_weights`
may be a `PartitionedVariable` as returned by using
`tf.compat.v1.get_variable()` with a
partitioner.
Invalid IDs (< 0) are pruned from input IDs and weights, as well as any IDs
with non-positive weight. For an entry with no features, the embedding vector
for `default_id` is returned, or the 0-vector if `default_id` is not supplied.
The ids and weights may be multi-dimensional. Embeddings are always aggregated
along the last dimension.
Args:
embedding_weights: A single tensor representing the complete embedding
tensor, or a list tensors all of same shape except for the first
dimension, representing sharded embedding tensors. Alternatively, a
`PartitionedVariable`, created by partitioning along dimension 0. Each
element must be appropriately sized for the given `partition_strategy`.
sparse_ids: `SparseTensor` of shape `[d_0, d_1, ..., d_n]` containing the
ids. `d_0` is typically batch size.
sparse_weights: `SparseTensor` of same shape as `sparse_ids`, containing
float weights corresponding to `sparse_ids`, or `None` if all weights are
be assumed to be 1.0.
combiner: A string specifying how to combine embedding results for each
entry. Currently "mean", "sqrtn" and "sum" are supported, with "mean" the
default.
default_id: The id to use for an entry with no features.
name: A name for this operation (optional).
partition_strategy: A string specifying the partitioning strategy. Currently
`"div"` and `"mod"` are supported. Default is `"div"`.
max_norm: If not `None`, all embeddings are l2-normalized to max_norm before
combining.
Returns:
A dense tensor representing the combined embeddings for the
sparse ids. For each row in the dense tensor represented by `sp_ids`, the op
looks up the embeddings for all ids in that row, multiplies them by the
corresponding weight, and combines these embeddings as specified.
In other words, if
`shape(combined embedding_weights) = [p0, p1, ..., pm]`
and
`shape(sparse_ids) = shape(sparse_weights) = [d0, d1, ..., dn]`
then
`shape(output) = [d0, d1, ... dn-1, p1, ..., pm]`.
For instance, if params is a 10x20 matrix, and sp_ids / sp_weights are
```python
[0, 0]: id 1, weight 2.0
[0, 1]: id 3, weight 0.5
[1, 0]: id -1, weight 1.0
[2, 3]: id 1, weight 3.0
```
`default_id` is 0.
with `combiner`="mean", then the output will be a 3x20 matrix where
```python
output[0, :] = (params[1, :] * 2.0 + params[3, :] * 0.5) / (2.0 + 0.5)
output[1, :] = (params[0, :] * 1.0) / 1.0
output[2, :] = (params[1, :] * 3.0) / 3.0
```
Raises:
ValueError: if `embedding_weights` is empty.
"""
if embedding_weights is None:
raise ValueError("Missing embedding_weights %s." % embedding_weights)
if isinstance(embedding_weights, variables.PartitionedVariable):
embedding_weights = list(embedding_weights) # get underlying Variables.
if not isinstance(embedding_weights, list):
embedding_weights = [embedding_weights]
if len(embedding_weights) < 1:
raise ValueError("Missing embedding_weights %s." % embedding_weights)
dtype = sparse_weights.dtype if sparse_weights is not None else None
embedding_weights = [
w if (isinstance(w, resource_variable_ops.ResourceVariable)
and dtype in (None, w.dtype))
else ops.convert_to_tensor(w, dtype=dtype)
for w in embedding_weights
]
with ops.name_scope(name, "embedding_lookup", embedding_weights +
[sparse_ids, sparse_weights]) as scope:
# Reshape higher-rank sparse ids and weights to linear segment ids.
original_shape = sparse_ids.dense_shape
original_rank_dim = tensor_shape.dimension_value(
sparse_ids.dense_shape.get_shape()[0])
original_rank = (
array_ops.size(original_shape)
if original_rank_dim is None else original_rank_dim)
sparse_ids = sparse_ops.sparse_reshape(sparse_ids, [
math_ops.reduce_prod(
array_ops.slice(original_shape, [0], [original_rank - 1])),
array_ops.gather(original_shape, original_rank - 1)
])
if sparse_weights is not None:
sparse_weights = sparse_tensor.SparseTensor(sparse_ids.indices,
sparse_weights.values,
sparse_ids.dense_shape)
# Prune invalid ids and weights.
sparse_ids, sparse_weights = _prune_invalid_ids(sparse_ids, sparse_weights)
if combiner != "sum":
sparse_ids, sparse_weights = _prune_invalid_weights(
sparse_ids, sparse_weights)
# Fill in dummy values for empty features, if necessary.
sparse_ids, is_row_empty = sparse_ops.sparse_fill_empty_rows(
sparse_ids, default_id or 0)
if sparse_weights is not None:
sparse_weights, _ = sparse_ops.sparse_fill_empty_rows(sparse_weights, 1.0)
result = embedding_lookup_sparse(
embedding_weights,
sparse_ids,
sparse_weights,
combiner=combiner,
partition_strategy=partition_strategy,
name=None if default_id is None else scope,
max_norm=max_norm)
if default_id is None:
# Broadcast is_row_empty to the same shape as embedding_lookup_result,
# for use in Select.
is_row_empty = array_ops.tile(
array_ops.reshape(is_row_empty, [-1, 1]),
array_ops.stack([1, array_ops.shape(result)[1]]))
result = array_ops.where(
is_row_empty, array_ops.zeros_like(result), result, name=scope)
# Reshape back from linear ids back into higher-dimensional dense result.
final_result = array_ops.reshape(
result,
array_ops.concat([
array_ops.slice(
math_ops.cast(original_shape, dtypes.int32), [0],
[original_rank - 1]),
array_ops.slice(array_ops.shape(result), [1], [-1])
], 0))
final_result.set_shape(
tensor_shape.unknown_shape(
(tensor_shape.Dimension(original_rank_dim) - 1).value).concatenate(
result.get_shape()[1:]))
return final_result
def testBroadcast_unknown_dims(self):
unknown = tensor_shape.unknown_shape()
shape_0 = tensor_shape.scalar()
shape_1 = tensor_shape.vector(1)
# pylint: disable=invalid-name
shape_U = tensor_shape.vector(None)
shape_1xU = tensor_shape.matrix(1, None)
shape_Ux1 = tensor_shape.matrix(None, 1)
shape_4xU = tensor_shape.matrix(4, None)
shape_Ux4 = tensor_shape.matrix(None, 4)
# pylint: enable=invalid-name
# Tensors with same shape should have the same broadcast result.
for shape in (shape_U, shape_1xU, shape_Ux1, shape_4xU, shape_Ux4):
self._assert_broadcast_with_unknown_dims(expected=shape,
shape1=shape,
shape2=shape)
# [] and [1] act like identity.
for identity in (shape_0, shape_1):
for shape in (shape_U, shape_1xU, shape_Ux1, shape_4xU, shape_Ux4):
self._assert_broadcast_with_unknown_dims(expected=shape,
shape1=identity,
shape2=shape)
# Unknown in, unknown out.
for shape in (shape_U, shape_1xU, shape_Ux1, shape_4xU, shape_Ux4):
self._assert_broadcast_with_unknown_dims(expected=unknown,
shape1=shape,
shape2=unknown)
self._assert_broadcast_with_unknown_dims(expected=shape_1xU,
shape1=shape_U,
shape2=shape_1xU)
shape_UxU = tensor_shape.matrix(None, None) # pylint: disable=invalid-name
self._assert_broadcast_with_unknown_dims(expected=shape_UxU,
shape1=shape_U,
shape2=shape_Ux1)
self._assert_broadcast_with_unknown_dims(expected=shape_4xU,
shape1=shape_U,
shape2=shape_4xU)
self._assert_broadcast_with_unknown_dims(expected=shape_Ux4,
shape1=shape_U,
shape2=shape_Ux4)
self._assert_broadcast_with_unknown_dims(expected=shape_UxU,
shape1=shape_1xU,
shape2=shape_Ux1)
self._assert_broadcast_with_unknown_dims(expected=shape_4xU,
shape1=shape_1xU,
shape2=shape_4xU)
self._assert_broadcast_with_unknown_dims(expected=shape_Ux4,
shape1=shape_1xU,
shape2=shape_Ux4)
self._assert_broadcast_with_unknown_dims(expected=shape_4xU,
shape1=shape_Ux1,
shape2=shape_4xU)
self._assert_broadcast_with_unknown_dims(expected=shape_Ux4,
shape1=shape_Ux1,
shape2=shape_Ux4)
shape_4x4 = tensor_shape.matrix(4, 4)
self._assert_broadcast_with_unknown_dims(expected=shape_4x4,
shape1=shape_4xU,
shape2=shape_Ux4)
