pytorch—torch.tensor.scatter操作解析

torch.Tensor.scatter_(dim, index, src, reduce&＃61;None)


理解scatter操作:
tensor_A.scatter_(dim, index, tensor_B): tensor_B的每个元素&＃xff0c;都按照 index 被scatter&＃xff08;可以理解为填充&＃xff09;到目标tensor_A中。

(1) index和源tensor_B维度一致;

(2) tensor_A一般是全零的张量&＃xff0c;其某些特定位置的值由 tensor_B 中的值填充。

(3) 注意如何根据index选取tensor_B中的值&＃xff1a;

对于2-D tensor:

if dim&＃61;0, tensor_A[index[i][j]][j] &＃61; tensor_B[i][j];if dim&＃61;1, tensor_A[i][index[i][j]] &＃61; tensor_B[i][j];


对于3-D tensor&＃xff1a;

if dim &＃61; 0&＃xff0c;tensor_A[index[i][j][k]][j][k] &＃61; tensor_B[i][j][k]
if dim &＃61; 1&＃xff0c;tensor_A[i][index[i][j][k]][k] &＃61; tensor_B[i][j][k]
if dim &＃61; 2&＃xff0c;tensor_A[i][j][index[i][j][k]] &＃61; tensor_B[i][j][k]


举例&＃xff1a;

如果对您有帮助&＃xff0c;麻烦点赞关注&＃xff0c;这真的对我很重要&＃xff01;&＃xff01;&＃xff01;如果需要互关&＃xff0c;请评论或者私信&＃xff01;