比较与torch.nn.NLLLoss的功能差异

查看源文件

torch.nn.NLLLoss

torch.nn.NLLLoss(
    weight=None,
    size_average=None,
    ignore_index=-100,
    reduce=None,
    reduction='mean'
)

更多内容详见torch.nn.NLLLoss

mindspore.ops.NLLLoss

class mindspore.ops.NLLLoss(
    reduction='mean'
)(logits, labels, weight)

更多内容详见mindspore.ops.NLLLoss

使用方式

PyTorch:同时支持二维数据 (N, C) 和多维数据(N, C, d1, d2, …, dK)。

MindSpore:仅支持二维数据 (N, C)。

迁移建议:如需要处理高维度输入数据,可以自行封装将d1, d2, …, dK维度拆分计算loss后再拼接的NLLLoss接口。

代码示例

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

# In MindSpore
m = nn.LogSoftmax(axis=1)
loss = ops.NLLLoss()
input = ms.Tensor(np.random.randn(3, 5), ms.float32)
labels = ms.Tensor([1, 0, 4], ms.int32)
weight = ms.Tensor(np.random.rand(5), ms.float32)
loss, weight = loss(m(input), labels, weight)
print(loss)
# Out:
# 1.3557988


# In PyTorch
m = torch.nn.LogSoftmax(dim=1)
loss = torch.nn.NLLLoss()
input = torch.randn(3, 5, requires_grad=True)
target = torch.tensor([1, 0, 4])
output = loss(m(input), target)
output.backward()
print(output)
# Out:
# tensor(1.7451, grad_fn=<NllLossBackward>)