交叉熵损失函数学习

参考
https://blog.csdn.net/weixin_43508499/article/details/107714080
https://blog.csdn.net/Lucinda6/article/details/116162198

主要看定义，函数的输入和输出

手动计算

import numpy as np
# 自定义numpy 的 softmax

#https://blog.csdn.net/qq_45759229/article/details/123583466
def soft_max(z):
    t = np.exp(z)
    a = np.exp(z) / np.sum(t, axis=1).reshape(-1,1)
    return a

# 假设有三个样本，总共的类别是5个，例如假设x是神经网络最后一层的输出，不需要经过softmax
x = np.array([[0.093, 0.1939, -1.0649, 0.4476, -2.0769],
              [-1.8024, 0.3696, 0.7796, -1.0346, 0.473],
              [0.5593, -2.5067, -2.1275, 0.5548, -1.6639]])
y = np.array([1, 2, 3])
sm_x=soft_max(x)
print(sm_x)

loss = -1*(np.log(sm_x[0][1]) + np.log(sm_x[1][2])+np.log(sm_x[2][3]))/3.0
loss

结果如下

numpy 实现

import torch
import numpy as np
from torch.nn import functional as F
import torch.nn as nn

# 定义softmax函数
def softmax(x):
    return np.exp(x) / np.sum(np.exp(x))

# 利用numpy计算
def cross_entropy_np(x, y):
    x_softmax = [softmax(x[i]) for i in range(len(x))]
    x_log = [np.log(x_softmax[i][y[i]]) for i in range(len(y))]
    loss = - np.sum(x_log) / len(y)
    return loss


# 另外一种实现方式，y转成onehot形式，比较直观
def cross_entropy_np2(x, y):
    num_data, num_class = x.shape
    log_p = np.array([np.log(softmax(x[i])) for i in range(num_data)])
    y_onehot = np.eye(num_class)[y]
    loss = - np.sum(y_onehot * log_p) / num_data
    return round(loss, 4)

if __name__ == '__main__':
    # 假设有数据x, y
    x = np.array([[0.093, 0.1939, -1.0649, 0.4476, -2.0769],
                  [-1.8024, 0.3696, 0.7796, -1.0346, 0.473],
                  [0.5593, -2.5067, -2.1275, 0.5548, -1.6639]])
    
    
    
    y = np.array([1, 2, 3])
    print('numpy result: ', cross_entropy_np(x, y)) 
    print('numpy result2：',cross_entropy_np2(x,y))

结果如下

pytorch手动计算

import torch
import numpy as np
from torch.nn import functional as F
import torch.nn as nn

if __name__ == '__main__':
    # 假设有数据x, y
    x = np.array([[0.093, 0.1939, -1.0649, 0.4476, -2.0769],
                  [-1.8024, 0.3696, 0.7796, -1.0346, 0.473],
                  [0.5593, -2.5067, -2.1275, 0.5548, -1.6639]])
    
    y = np.array([1, 2, 3])
    x_tensor = torch.from_numpy(x)
    y_tensor = torch.from_numpy(y).long()
    
    softmax = nn.Softmax()
    x_softmax = softmax(x_tensor)

    #print(x_softmax)

    #x_softmax
    x_log = torch.log(x_softmax)

    #print(x_log)

    ls = nn.LogSoftmax(dim=1)
    #print(ls(x))

    loss = x_log[range(len(x)),y_tensor]  #取出每一个样本标签值处的概率
    loss = abs(sum(loss)/len(x))

    print(loss)

    loss_func = nn.NLLLoss()
    print(loss_func(x_log,y_tensor))

pytorch实现（使用nn.CrossEntropy或F.cross_entropy）

import torch
import numpy as np
from torch.nn import functional as F
import torch.nn as nn

if __name__ == '__main__':
    # 假设有数据x, y
    x = np.array([[0.093, 0.1939, -1.0649, 0.4476, -2.0769],
                  [-1.8024, 0.3696, 0.7796, -1.0346, 0.473],
                  [0.5593, -2.5067, -2.1275, 0.5548, -1.6639]])
    
    
    
    y = np.array([1, 2, 3])
    print('numpy result: ', cross_entropy_np(x, y)) 
    print('numpy result2：',cross_entropy_np2(x,y))
    x_tensor = torch.from_numpy(x)
    y_tensor = torch.from_numpy(y).long()
    cross_entropy_torch1 = nn.CrossEntropyLoss()
    print("pytorch result1:", cross_entropy_torch1(x_tensor,y_tensor))
    
    cross_entropy_torch2= F.cross_entropy
    print("pytorch result2:", cross_entropy_torch1(x_tensor,y_tensor))
    
    
    #### y 不是非得从0开始的，这个错误的，标签只需要是0-nclass-1即可
    x_tensor = torch.from_numpy(x)
    y_tensor = torch.from_numpy(y-1).long()
    cross_entropy_torch = nn.CrossEntropyLoss()
    print("pytorch false result:", cross_entropy_torch(x_tensor,y_tensor))
# numpy result:  1.0155949508195155

## 

结果如下