pytorch-10 神经网络的损失函数

1. 回归：SSE和MSE

# MSE损失函数
import torch
from torch.nn import MSELoss

yhat = torch.randn(size=(50,), dtype=torch.float32)
y = torch.randn(size=(50,), dtype=torch.float32)

criterion = MSELoss()
loss1 = criterion(yhat, y)

# 计算mse 误差平方
criterion = MSELoss(reduction="mean")
loss2 = criterion(yhat, y)

# 计算sse
criterion = MSELoss(reduction="sum")
loss3 = criterion(yhat, y)

loss1, loss2, loss3

2. BCELoss二分类交叉熵损失函数

方法1：nn模块中的类

• class BCEWithLogitsLoss
• class BCELoss

方法2：functional库中的计算函数 （很少用到）

• function F.binary_cross_entropy_with_logits
• function F.binary_cross_entropy

# 二分类交叉熵损失函数 方法一、手动实现 BCELoss
import torch

N = 3*pow(10,3)
torch.random.manual_seed(420)
X = torch.rand((N,4),dtype=torch.float32)
w = torch.rand((4,1),dtype=torch.float32,requires_grad=True)
y = torch.randint(low=0,high=2,size=(N,1),dtype=torch.float32)

zhat = torch.mm(X,w)
sigma = torch.sigmoid(zhat)

loss = -(1/N) * torch.sum( (1-y)*torch.log(1-sigma) + y*torch.log(sigma) )  # binary cross entropy loss
loss

# BCELoss 方法二、使用类
import torch
import torch.nn as nn

# X, w, y
# zhat, sigma(sigmoid)

criterion = nn.BCELoss()  # 不带sigmoid函数, 主要为了监控准确率
loss1 = criterion(sigma, y)

criterion = nn.BCEWithLogitsLoss()  # 带有sigmoid函数
loss2 = criterion(zhat, y)

loss1, loss2

# BCELoss 方法三、使用函数
import torch
from torch.nn import functional as F

# X, w, y
# zhat, sigma(sigmoid)

loss1 = F.binary_cross_entropy(sigma, y) # 没有sigmoid函数
loss2 = F.binary_cross_entropy_with_logits(zhat, y) # 有sigmoid函数

loss1, loss2

3. CrossEntropyLoss 多分类交叉熵损失函数

方法1：调用logsoftmax和NLLLoss实现
方法2：直接调用CrossEntropyLoss

# 多分类交叉熵损失函数
import torch
import torch.nn as nn

N = 3*pow(10,2)
torch.random.manual_seed(420)
X = torch.rand((N,4),dtype=torch.float32)
w = torch.rand((4,3),dtype=torch.float32,requires_grad=True)
y = torch.randint(low=0,high=2,size=(N,),dtype=torch.float32)

# 方法一、LogSoftmax 和 NLLLoss
zhat = torch.mm(X,w)
logsm = nn.LogSoftmax(dim=1)
logsigma = logsm(zhat)

criterion = nn.NLLLoss() # 将标签转化成 独热编码，01的稀疏矩阵，类型是Long
loss1 = criterion(logsigma, y.long())

# 方法二、CrossEntropyLoss
criterion = nn.CrossEntropyLoss()  # 属性reduction的值，mean、sum、None
loss2 = criterion(zhat, y.long())

criterion = nn.CrossEntropyLoss(reduction="mean")  # 属性reduction的值，mean、sum、none
loss_mean = criterion(zhat, y.long())

criterion = nn.CrossEntropyLoss(reduction="sum")  # 属性reduction的值，mean、sum、none
loss_sum = criterion(zhat, y.long())

criterion = nn.CrossEntropyLoss(reduction="none") # 没有聚合函数的聚合效果
loss_none = criterion(zhat, y.long())

loss1, loss2, loss_mean, loss_sum, loss_none