gather_nd¶
该OP是 gather
的高维推广,并且支持多轴同时索引。 index
是一个K维度的张量,它可以认为是从 x
中取K-1维张量,每一个元素是一个切片:
\[output[(i_0, ..., i_{K-2})] = x[index[(i_0, ..., i_{K-2})]]\]
显然, index.shape[-1] <= x.rank
并且输出张量的维度是 index.shape[:-1] + x.shape[index.shape[-1]:]
。
示例:
给定:
x = [[[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]],
[[12, 13, 14, 15],
[16, 17, 18, 19],
[20, 21, 22, 23]]]
x.shape = (2, 3, 4)
- 案例 1:
index = [[1]]
gather_nd(x, index)
= [x[1, :, :]]
= [[12, 13, 14, 15],
[16, 17, 18, 19],
[20, 21, 22, 23]]
- 案例 2:
index = [[0,2]]
gather_nd(x, index)
= [x[0, 2, :]]
= [8, 9, 10, 11]
- 案例 3:
index = [[1, 2, 3]]
gather_nd(x, index)
= [x[1, 2, 3]]
= [23]
- 参数:
-
x (Tensor) - 输入Tensor,数据类型可以是int32,int64,float32,float64, bool。
index (Tensor) - 输入的索引Tensor,其数据类型int32或者int64。它的维度
index.rank
必须大于1,并且index.shape[-1] <= x.rank
。name (str,可选)- 具体用法请参见 Name ,一般无需设置,默认值为None。
返回:shape为index.shape[:-1] + x.shape[index.shape[-1]:]的Tensor,数据类型与 x
一致。
代码示例:
import paddle
import numpy as np
np_x = np.array([[[1, 2], [3, 4], [5, 6]],
[[7, 8], [9, 10], [11, 12]]])
np_index = [[0, 1]]
x = paddle.to_tensor(np_x)
index = paddle.to_tensor(np_index)
output = paddle.gather_nd(x, index) #[[3, 4]]