比较与torch.Tensor.scatter_add_的功能差异

查看源文件

torch.Tensor.scatter_add_

torch.Tensor.scatter_add_(
    dim,
    index,
    src
)

更多内容详见torch.Tensor.scatter_add_

mindspore.ops.ScatterNdAdd

class mindspore.ops.ScatterNdAdd(use_locking=False)(
    input_x,
    indices,
    update
)

更多内容详见mindspore.ops.ScatterNdAdd

使用方式

PyTorch:给定输入tensor,更新tensor和索引tensor;将更新tensor按照索引tensor在指定的轴上加到输入tensor上。

MindSpore:给定输入tensor,更新tensor和索引tensor;将更新tensor按照索引tensor加到输入tensor上; 不支持通过参数自定义轴,但可通过调整索引tensor的形状来明确轴。

代码示例

import mindspore as ms
import mindspore.ops as ops
import torch
import numpy as np

# In MindSpore, no parameter for specifying dimension.
input_x = ms.Parameter(ms.Tensor(np.array([1, 2, 3, 4, 5, 6, 7, 8]), ms.float32), name="x")
indices = ms.Tensor(np.array([[2], [4], [1], [7]]), mindspore.int32)
updates = ms.Tensor(np.array([6, 7, 8, 9]), ms.float32)
scatter_nd_add = ops.ScatterNdAdd()
output = scatter_nd_add(input_x, indices, updates)
print(output)
# Out:
# [1. 10. 9. 4. 12. 6. 7. 17.]

# In torch, parameter dim can be set to specify dimension.
input_x = torch.tensor(np.array([1, 2, 3, 4, 5, 6, 7, 8]).astype(np.float32))
indices = torch.tensor(np.array([2, 4, 1, 7]).astype(np.int64))
updates = torch.tensor(np.array([6, 7, 8, 9]).astype(np.float32))
output = input_x.scatter_add_(dim=0, index=indices, src=updates)
print(output)
# Out:
# tensor([1., 10., 9., 4., 12., 6., 7., 17.])