# gather_nd¶

paddle.fluid.layers.nn. gather_nd ( input, index, name=None ) [source]

Gather Nd Layer

This function is actually a high-dimensional extension of `gather` and supports for simultaneous indexing by multiple axes. `index` is a K-dimensional integer tensor, which is regarded as a (K-1)-dimensional tensor of `index` into `input`, where each element defines a slice of params:

\[output[(i_0, ..., i_{K-2})] = input[index[(i_0, ..., i_{K-2})]]\]

Obviously, `index.shape[-1] <= input.rank` . And, the output tensor has shape `index.shape[:-1] + input.shape[index.shape[-1]:]` .

```Given:
input = [[[ 0,  1,  2,  3],
[ 4,  5,  6,  7],
[ 8,  9, 10, 11]],
[[12, 13, 14, 15],
[16, 17, 18, 19],
[20, 21, 22, 23]]]
input.shape = (2, 3, 4)

* Case 1:
index = [[1]]

gather_nd(input, index)
= [input[1, :, :]]
= [[12, 13, 14, 15],
[16, 17, 18, 19],
[20, 21, 22, 23]]

* Case 2:
index = [[0,2]]

gather_nd(input, index)
= [input[0, 2, :]]
= [8, 9, 10, 11]

* Case 3:
index = [[1, 2, 3]]

gather_nd(input, index)
= [input[1, 2, 3]]
= [23]
```
Parameters
• input (Tensor) – The input Tensor which it’s data type should be bool, float32, float64, int32, int64.

• index (Tensor) – The index input with rank > 1, index.shape[-1] <= input.rank. Its dtype should be int32, int64.

• name (str, optional) – The default value is None. Normally there is no need for user to set this property. For more information, please refer to Name .

Returns

A tensor with the shape index.shape[:-1] + input.shape[index.shape[-1]:]

Return type

output (Tensor)

Examples

```import paddle
import paddle.fluid as fluid