T
- data type for output()
output@Operator public final class TensorScatterAdd<T> extends PrimitiveOp implements Operand<T>
This operation creates a new tensor by adding sparse `updates` to the passed in `tensor`. This operation is very similar to `tf.scatter_nd_add`, except that the updates are added onto an existing tensor (as opposed to a variable). If the memory for the existing tensor cannot be re-used, a copy is made and updated.
`indices` is an integer tensor containing indices into a new tensor of shape `shape`. The last dimension of `indices` can be at most the rank of `shape`:
indices.shape[-1] <= shape.rank
The last dimension of `indices` corresponds to indices into elements (if `indices.shape[-1] = shape.rank`) or slices (if `indices.shape[-1] < shape.rank`) along dimension `indices.shape[-1]` of `shape`. `updates` is a tensor with shape
indices.shape[:-1] + shape[indices.shape[-1]:]
The simplest form of tensor_scatter_add is to add individual elements to a tensor by index. For example, say we want to add 4 elements in a rank-1 tensor with 8 elements.
In Python, this scatter add operation would look like this:
indices = tf.constant([[4], [3], [1], [7]])
updates = tf.constant([9, 10, 11, 12])
tensor = tf.ones([8], dtype=tf.int32)
updated = tf.tensor_scatter_add(tensor, indices, updates)
with tf.Session() as sess:
print(sess.run(scatter))
The resulting tensor would look like this:
[1, 12, 1, 11, 10, 1, 1, 13]
We can also, insert entire slices of a higher rank tensor all at once. For example, if we wanted to insert two slices in the first dimension of a rank-3 tensor with two matrices of new values.
In Python, this scatter add operation would look like this:
indices = tf.constant([[0], [2]])
updates = tf.constant([[[5, 5, 5, 5], [6, 6, 6, 6],
[7, 7, 7, 7], [8, 8, 8, 8]],
[[5, 5, 5, 5], [6, 6, 6, 6],
[7, 7, 7, 7], [8, 8, 8, 8]]])
tensor = tf.ones([4, 4, 4])
updated = tf.tensor_scatter_add(tensor, indices, updates)
with tf.Session() as sess:
print(sess.run(scatter))
The resulting tensor would look like this:
[[[6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8], [9, 9, 9, 9]], [[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]], [[6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8], [9, 9, 9, 9]], [[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]]]
Note that on CPU, if an out of bound index is found, an error is returned. On GPU, if an out of bound index is found, the index is ignored.
operation
Modifier and Type | Method and Description |
---|---|
Output<T> |
asOutput()
Returns the symbolic handle of a tensor.
|
static <T,U extends Number> |
create(Scope scope,
Operand<T> tensor,
Operand<U> indices,
Operand<T> updates)
Factory method to create a class wrapping a new TensorScatterAdd operation.
|
Output<T> |
output()
A new tensor copied from tensor and updates added according to the indices.
|
equals, hashCode, op, toString
public static <T,U extends Number> TensorScatterAdd<T> create(Scope scope, Operand<T> tensor, Operand<U> indices, Operand<T> updates)
scope
- current scopetensor
- Tensor to copy/update.indices
- Index tensor.updates
- Updates to scatter into output.public Output<T> output()
public Output<T> asOutput()
Operand
Inputs to TensorFlow operations are outputs of another TensorFlow operation. This method is used to obtain a symbolic handle that represents the computation of the input.
asOutput
in interface Operand<T>
OperationBuilder.addInput(Output)
Copyright © 2022. All rights reserved.