def sparse_read(self, indices, collections=None, trainable=True, name=None):
"""Reads the value of this variable sparsely, using `gather`."""
with ops.name_scope("Gather" if name is None else name):
value = gen_resource_variable_ops.resource_gather(
self._handle, indices, dtype=self._dtype)
_register_variable_read(value, collections=collections, trainable=trainable)
return array_ops.identity(value)
def sparse_read(self, indices, name=None):
"""Reads the value of this variable sparsely, using `gather`."""
with ops.name_scope("Gather" if name is None else name) as name:
if self._trainable:
tape.watch_variable(self)
value = gen_resource_variable_ops.resource_gather(
self._handle, indices, dtype=self._dtype, name=name)
return array_ops.identity(value)
def sparse_read(self, indices, name=None):
"""Reads the value of this variable sparsely, using `gather`."""
with ops.name_scope("Gather" if name is None else name) as name:
if self._trainable:
tape.watch_variable(self)
value = gen_resource_variable_ops.resource_gather(
self._handle, indices, dtype=self._dtype, name=name)
return array_ops.identity(value)
def sparse_read(self,
indices,
collections=None,
trainable=True,
name=None):
with ops.name_scope("Gather" if name is None else name):
value = gen_resource_variable_ops.resource_gather(
self._handle, indices, dtype=self._dtype)
_register_variable_read(value,
collections=collections,
trainable=trainable)
return value
def resource_gather(resource,
indices,
dtype,
validate_indices=True,
name=None):
"""Gather slices from the variable pointed to by `resource`.
`indices` must be an integer tensor of any dimension (usually 0-D or 1-D).
Produces an output tensor with shape `indices.shape + params.shape[1:]` where:
```python
# Scalar indices
output[:, ..., :] = params[indices, :, ... :]
# Vector indices
output[i, :, ..., :] = params[indices[i], :, ... :]
# Higher rank indices
output[i, ..., j, :, ... :] = params[indices[i, ..., j], :, ..., :]
```
Args:
resource: A `Tensor` of type `resource`.
handle to the resource in which to store the variable.
indices: a integer `Tensor` containing the indices to be gathered.
dtype: A `tf.DType`. the dtype of the value.
validate_indices: optional `bool`. If false will not validate that the
indices fit in the variable.
name: The optional name for the operation to be added.
Returns:
A `Tensor` of type `dtype`.
"""
result = gen_resource_variable_ops.resource_gather(
resource, indices, dtype, validate_indices=validate_indices, name=name)
def grad(dresult):
return ops.IndexedSlices(dresult,
indices,
dense_shape=ops.convert_to_tensor(
resource._variable_shape)) # pylint: disable=protected-access
return result, grad
def resource_gather(resource, indices, dtype, validate_indices=True, name=None):
"""Gather slices from the variable pointed to by `resource`.
`indices` must be an integer tensor of any dimension (usually 0-D or 1-D).
Produces an output tensor with shape `indices.shape + params.shape[1:]` where:
```python
# Scalar indices
output[:, ..., :] = params[indices, :, ... :]
# Vector indices
output[i, :, ..., :] = params[indices[i], :, ... :]
# Higher rank indices
output[i, ..., j, :, ... :] = params[indices[i, ..., j], :, ..., :]
```
Args:
resource: A `Tensor` of type `resource`.
handle to the resource in which to store the variable.
indices: a integer `Tensor` containing the indices to be gathered.
dtype: A `tf.DType`. the dtype of the value.
validate_indices: optional `bool`. If false will not validate that the
indices fit in the variable.
name: The optional name for the operation to be added.
Returns:
A `Tensor` of type `dtype`.
"""
result = gen_resource_variable_ops.resource_gather(
resource, indices, dtype, validate_indices=validate_indices, name=name)
def grad(dresult):
return ops.IndexedSlices(
dresult,
indices,
dense_shape=gen_resource_variable_ops.variable_shape(resource))
return result, grad
def sparse_read(self, indices, collections=None, trainable=True, name=None):
with ops.name_scope("Gather" if name is None else name):
value = gen_resource_variable_ops.resource_gather(
self._handle, indices, dtype=self._dtype)
_register_variable_read(value, collections=collections, trainable=trainable)
return value
