比较与torch.nn.functional.mse_loss的功能差异

torch.nn.functional.mse_loss

torch.nn.functional.mse_loss(input, target, size_average=None, reduce=None, reduction='mean') -> Tensor

更多内容详见torch.nn.functional.mse_loss。

mindspore.nn.MSELoss

class mindspore.nn.MSELoss(reduction='mean')(logits, labels) -> Tensor

更多内容详见mindspore.nn.MSELoss。

差异对比

PyTorch：用于计算输入x和y每一个元素的均方误差，reduction参数指定应用于loss的规约类型。

MindSpore：实现与PyTorch一致的功能。

分类	子类	PyTorch	MindSpore	差异
参数	参数1	input	logits	功能一致，参数名不同
	参数2	target	labels	功能一致，参数名不同
	参数3	size_average	-	被reduction替代
	参数4	reduce	-	被reduction替代
	参数5	reduction	reduction	-

代码示例1

计算input和target的均方误差。

# PyTorch
import torch
from torch.nn.functional import mse_loss
from torch import tensor
import numpy as np

# 默认情况，reduction='mean'
input_ = np.array([1, 1, 1, 1]).reshape((2, 2))
input = tensor(input_, dtype=torch.float32)
target_ = np.array([1, 2, 2, 1]).reshape((2, 2))
target = tensor(target_, dtype=torch.float32)
output = mse_loss(input, target)
print(output.numpy())
# 0.5

# MindSpore
import mindspore
from mindspore import Tensor
import mindspore.nn as nn
import numpy as np

loss = nn.MSELoss()
input_ = np.array([1, 1, 1, 1]).reshape((2, 2))
input = Tensor(input_, dtype=mindspore.float32)
target_ = np.array([1, 2, 2, 1]).reshape((2, 2))
target = Tensor(target_, dtype=mindspore.float32)
output = loss(input, target)
print(output)
# 0.5

代码示例2

计算input和target的均方误差，以求和方式规约。

# PyTorch
import torch
from torch.nn.functional import mse_loss
from torch import tensor
import numpy as np

# redcution='sum'
input_ = np.array([1, 1, 1, 1]).reshape((2, 2))
input = tensor(input_, dtype=torch.float32)
target_ = np.array([1, 2, 2, 1]).reshape((2, 2))
target = tensor(target_, dtype=torch.float32)
print(target)
output = mse_loss(input, target, reduction='sum')
print(output.numpy())
# 2.0


# MindSpore
import mindspore
from mindspore import Tensor
import mindspore.nn as nn
import numpy as np

loss = nn.MSELoss(reduction='sum')
input_ = np.array([1, 1, 1, 1]).reshape((2, 2))
input = Tensor(input_, dtype=mindspore.float32)
target_ = np.array([1, 2, 2, 1]).reshape((2, 2))
target = Tensor(target_, dtype=mindspore.float32)
output = loss(input, target)
print(output)
# 2.0