Pytorch - scatter_()

torch_scatter.scatter(src: Tensor, index: Tensor, dim: int = - 1, 
    out: Optional[Tensor] = None,
     dim_size: Optional[int] = None,
     reduce: str = 'sum')→ Tensor

• src – The source tensor. ( 源张量)

• index – The indices of elements to scatter.(要分散的元素的索引)

• dim – The axis along which to index. (default: -1) (要索引的轴（默认值：-1）)

• out – The destination tensor. (目标张量)

• dim_size – If out is not given, automatically create output with size dim_size at dimension dim. If dim_size is not given, a minimal sized output tensor according to index.max() + 1 is returned.(如果未给出out，则在 dim 处自动创建尺寸为 dim_size 的输出。如果没有给出 dim_size，则返回根据 index.max() + 1 的最小尺寸输出张量);

• reduce – The reduce operation ("sum", "mul", "mean", "min" or "max"). (default: "sum") (reduce 操作(“ sum”、“ mul”、“ mean”、“ min”或“ max”);

直观的理解:

对于三维矩阵：

y = y.scatter(dim,index,src)
 
#则结果为：
y[ index[i][j][k]  ] [j][k] = src[i][j][k] # if dim == 0
y[i] [ index[i][j][k] ] [k] = src[i][j][k] # if dim == 1
y[i][j] [ index[i][j][k] ]  = src[i][j][k] # if dim == 2

对于二维矩阵：

y = y.scatter(dim,index,src)
 
#则：
y [ index[i][j] ] [j] = src[i][j] #if dim==0
y[i] [ index[i][j] ]  = src[i][j] #if dim==1 

ps: index的维度，必须和src维度相同；

举例：

>>> src = torch.randn(3, 3)
>>> src
tensor([[-1.8801,  0.9740,  1.2865],
        [ 0.3140,  1.2396, -1.3452],
        [-0.8937,  0.6916, -2.0134]])
>>> y = y.scatter_(0,index,src)
>>> index = torch.tensor([[0, 1, 0],[1,0,1],[2,1,0]])
>>> index
tensor([[0, 1, 0],
        [1, 0, 1],
        [2, 1, 0]])
>>> y = y.scatter_(0,index,src)
>>> y
tensor([[-1.8801,  1.2396, -2.0134],
        [ 0.3140,  0.6916, -1.3452],
        [-0.8937,  0.0000,  0.0000],
        [ 0.0000,  0.0000,  0.0000]])