交叉熵+全連接層的構建

1. 求熵

import torch

def Entropy(x):
    y = -(x * torch.log2(x)).sum()
    return y

a = torch.full([4],1/4)
b = torch.tensor([0.1, 0.1, 0.1, 0.7])
c = torch.tensor([0.001, 0.001, 0.001, 0.999])

if __name__ == "__main__":
    y1 = Entropy(a)
    y2 = Entropy(b)
    y3 = Entropy(c)

    print(y1, y2, y3)

結果：

tensor(2.) tensor(1.3568) tensor(0.0313)

2. 交叉熵

import torch
import torch.nn.functional as F

x = torch.randn(1,784)
w = torch.randn(10,784)

logits = x@w.t()
pred = F.softmax(logits, dim=1)
pred_log = torch.log(pred)

print(F.cross_entropy(logits, torch.tensor([3])))
print(F.nll_loss(pred_log,torch.tensor([3])))

tensor(28.7205)
tensor(28.7205)

3.用交叉熵進行多分類問題

import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import datasets, transforms

batch_size = 200
learning_rate = 0.01
epochs = 10

train_loader = torch.utils.data.DataLoader(
    datasets.MNIST('../data', train=True,
                   transform=transforms.Compose([
                       transforms.ToTensor(),
                       transforms.Normalize((0.1307,), (0.3081,))
                   ])),
    batch_size=batch_size, shuffle=True)
test_loader = torch.utils.data.DataLoader(
    datasets.MNIST('../data', train=False, transform=transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize((0.1307,), (0.3081,))
    ])),
    batch_size=batch_size, shuffle=True)

w1, b1 = torch.randn(200, 784, requires_grad=True), \
         torch.zeros(200, requires_grad=True)
w2, b2 = torch.randn(200, 200, requires_grad=True), \
         torch.zeros(200, requires_grad=True)
w3, b3 = torch.randn(10, 200, requires_grad=True), \
         torch.zeros(10, requires_grad=True)

##初始化
torch.nn.init.kaiming_normal_(w1)
torch.nn.init.kaiming_normal_(w2)
torch.nn.init.kaiming_normal_(w3)


def forward(x):
    x = x @ w1.t() + b1
    x = F.relu(x)
    x = x @ w2.t() + b2
    x = F.relu(x)
    x = x @ w3.t() + b3
    x = F.relu(x)
    return x

optimizer = torch.optim.SGD([w1, b1, w2, b2, w3, b3], lr=learning_rate)
criteon = nn.CrossEntropyLoss()

for epoch in range(epochs):

    for batch_idx, (data, target) in enumerate(train_loader):
        data = data.view(-1, 28 * 28)

        logits = forward(data)
        loss = criteon(logits, target)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if batch_idx % 100 == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset),
                       100. * batch_idx / len(train_loader), loss.item()))

    test_loss = 0
    correct = 0
    for data, target in test_loader:
        data = data.view(-1, 28 * 28)
        logits = forward(data)
        test_loss += criteon(logits, target).item()

        pred = logits.data.max(1)[1]
        correct += pred.eq(target.data).sum()

    test_loss /= len(test_loader.dataset)
    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        100. * correct / len(test_loader.dataset)))

運行結果：

Train Epoch: 0 [0/60000 (0%)]   Loss: 2.816831
Train Epoch: 0 [20000/60000 (33%)]      Loss: 0.695306
Train Epoch: 0 [40000/60000 (67%)]      Loss: 0.550181

Test set: Average loss: 0.0018, Accuracy: 8973/10000 (89%)

Train Epoch: 1 [0/60000 (0%)]   Loss: 0.334501
Train Epoch: 1 [20000/60000 (33%)]      Loss: 0.430899
Train Epoch: 1 [40000/60000 (67%)]      Loss: 0.378964

Test set: Average loss: 0.0014, Accuracy: 9215/10000 (92%)

Train Epoch: 2 [0/60000 (0%)]   Loss: 0.299217
Train Epoch: 2 [20000/60000 (33%)]      Loss: 0.254395
Train Epoch: 2 [40000/60000 (67%)]      Loss: 0.297133

Test set: Average loss: 0.0012, Accuracy: 9307/10000 (93%)

Train Epoch: 3 [0/60000 (0%)]   Loss: 0.231520
Train Epoch: 3 [20000/60000 (33%)]      Loss: 0.220551
Train Epoch: 3 [40000/60000 (67%)]      Loss: 0.214800

Test set: Average loss: 0.0011, Accuracy: 9398/10000 (93%)

Train Epoch: 4 [0/60000 (0%)]   Loss: 0.203065
Train Epoch: 4 [20000/60000 (33%)]      Loss: 0.252411
Train Epoch: 4 [40000/60000 (67%)]      Loss: 0.227913

Test set: Average loss: 0.0010, Accuracy: 9442/10000 (94%)

Train Epoch: 5 [0/60000 (0%)]   Loss: 0.197479
Train Epoch: 5 [20000/60000 (33%)]      Loss: 0.155061
Train Epoch: 5 [40000/60000 (67%)]      Loss: 0.231575

Test set: Average loss: 0.0009, Accuracy: 9470/10000 (94%)

Train Epoch: 6 [0/60000 (0%)]   Loss: 0.139516
Train Epoch: 6 [20000/60000 (33%)]      Loss: 0.182578
Train Epoch: 6 [40000/60000 (67%)]      Loss: 0.189277

Test set: Average loss: 0.0009, Accuracy: 9512/10000 (95%)

Train Epoch: 7 [0/60000 (0%)]   Loss: 0.231095
Train Epoch: 7 [20000/60000 (33%)]      Loss: 0.137316
Train Epoch: 7 [40000/60000 (67%)]      Loss: 0.159869

Test set: Average loss: 0.0008, Accuracy: 9541/10000 (95%)

Train Epoch: 8 [0/60000 (0%)]   Loss: 0.164737
Train Epoch: 8 [20000/60000 (33%)]      Loss: 0.082444
Train Epoch: 8 [40000/60000 (67%)]      Loss: 0.103773

Test set: Average loss: 0.0008, Accuracy: 9560/10000 (95%)

Train Epoch: 9 [0/60000 (0%)]   Loss: 0.139510
Train Epoch: 9 [20000/60000 (33%)]      Loss: 0.100558
Train Epoch: 9 [40000/60000 (67%)]      Loss: 0.188554

Test set: Average loss: 0.0007, Accuracy: 9576/10000 (95%)

4.用nn.Linear構建全連接層

import torch
import torch.nn as nn
import torch.nn.functional as F
x = torch.randn(1,784)

layer1 = nn.Linear(784,200) #in到out
layer2 = nn.Linear(200,200)
layer3 = nn.Linear(200,10)

x = layer1(x)
x = F.relu(x,inplace=True)  #保持非線性
x = layer2(x)
x = F.relu(x,inplace=True)
x = layer3(x)
x = F.relu(x,inplace=True)  

print(x.shape)

torch.Size([1, 10])

5. 繼承nn.Module構建全連接層

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets,transforms

batch_size = 200
epochs = 10
learning_rate = 0.01

train_loader = torch.utils.data.DataLoader(
    datasets.MNIST('../data', train=True,
                   transform=transforms.Compose([
                       transforms.ToTensor(),
                       transforms.Normalize((0.1307,), (0.3081,))
                   ])),
    batch_size=batch_size, shuffle=True)
test_loader = torch.utils.data.DataLoader(
    datasets.MNIST('../data', train=False, transform=transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize((0.1307,), (0.3081,))
    ])),
    batch_size=batch_size, shuffle=True) 



class MLP(nn.Module):  #要繼承nn.Module
    
    def __init__(self):  #初始化,可自帶參數(這裏沒帶，因爲已給出）
        super(MLP, self).__init__()

        self.model = nn.Sequential(
            nn.Linear(784,200),
            nn.ReLU(inplace=True),
            nn.Linear(200,200),
            nn.ReLU(inplace=True),
            nn.Linear(200,10),
            nn.ReLU(inplace=True),
        )

    def forward(self,x):
        x = self.model(x)

        return x


#train
device = torch.device('cuda:0')
net = MLP().to(device)

optimizer = optim.SGD(net.parameters(),lr=learning_rate)
criteon = nn.CrossEntropyLoss()

for epoch in range(epochs):

    for batch_idx, (data, target) in enumerate(train_loader):
        data = data.view(-1, 28 * 28)
        data, target = data.to(device), target.cuda()

        logits = net(data)
        loss = criteon(logits, target)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if batch_idx % 100 == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset),
                       100. * batch_idx / len(train_loader), loss.item()))

    test_loss = 0
    correct = 0
    for data, target in test_loader:
        data = data.view(-1, 28 * 28)
        data, target = data.to(device), target.cuda()
        logits = net(data)
        test_loss += criteon(logits, target).item()

        pred = logits.data.max(1)[1]
        correct += pred.eq(target.data).sum()

    test_loss /= len(test_loader.dataset)
    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        100. * correct / len(test_loader.dataset)))

Train Epoch: 0 [0/60000 (0%)]   Loss: 2.309051
Train Epoch: 0 [20000/60000 (33%)]      Loss: 2.110557
Train Epoch: 0 [40000/60000 (67%)]      Loss: 1.602919

Test set: Average loss: 0.0057, Accuracy: 7220/10000 (72%)

Train Epoch: 1 [0/60000 (0%)]   Loss: 1.223766
Train Epoch: 1 [20000/60000 (33%)]      Loss: 0.843435
Train Epoch: 1 [40000/60000 (67%)]      Loss: 0.826475

Test set: Average loss: 0.0034, Accuracy: 7994/10000 (79%)

Train Epoch: 2 [0/60000 (0%)]   Loss: 0.755618
Train Epoch: 2 [20000/60000 (33%)]      Loss: 0.544501
Train Epoch: 2 [40000/60000 (67%)]      Loss: 0.517594

Test set: Average loss: 0.0028, Accuracy: 8441/10000 (84%)

Train Epoch: 3 [0/60000 (0%)]   Loss: 0.611635
Train Epoch: 3 [20000/60000 (33%)]      Loss: 0.511144
Train Epoch: 3 [40000/60000 (67%)]      Loss: 0.603972

Test set: Average loss: 0.0026, Accuracy: 8738/10000 (87%)

Train Epoch: 4 [0/60000 (0%)]   Loss: 0.565730
Train Epoch: 4 [20000/60000 (33%)]      Loss: 0.513728
Train Epoch: 4 [40000/60000 (67%)]      Loss: 0.529109

Test set: Average loss: 0.0025, Accuracy: 8881/10000 (88%)

Train Epoch: 5 [0/60000 (0%)]   Loss: 0.525373
Train Epoch: 5 [20000/60000 (33%)]      Loss: 0.470926
Train Epoch: 5 [40000/60000 (67%)]      Loss: 0.517402

Test set: Average loss: 0.0024, Accuracy: 8978/10000 (89%)

Train Epoch: 6 [0/60000 (0%)]   Loss: 0.505295
Train Epoch: 6 [20000/60000 (33%)]      Loss: 0.457965
Train Epoch: 6 [40000/60000 (67%)]      Loss: 0.552664

Test set: Average loss: 0.0023, Accuracy: 9052/10000 (90%)

Train Epoch: 7 [0/60000 (0%)]   Loss: 0.635619
Train Epoch: 7 [20000/60000 (33%)]      Loss: 0.365278
Train Epoch: 7 [40000/60000 (67%)]      Loss: 0.370224

Test set: Average loss: 0.0023, Accuracy: 9117/10000 (91%)

Train Epoch: 8 [0/60000 (0%)]   Loss: 0.400364
Train Epoch: 8 [20000/60000 (33%)]      Loss: 0.452789
Train Epoch: 8 [40000/60000 (67%)]      Loss: 0.415130

Test set: Average loss: 0.0022, Accuracy: 9188/10000 (91%)

Train Epoch: 9 [0/60000 (0%)]   Loss: 0.464027
Train Epoch: 9 [20000/60000 (33%)]      Loss: 0.385774
Train Epoch: 9 [40000/60000 (67%)]      Loss: 0.410523

Test set: Average loss: 0.0022, Accuracy: 9183/10000 (91%)

可以在程序運行時，打來任務管理器看GPU的使用情況：

交叉熵+全連接層的構建

目錄

1. 求熵

2. 交叉熵

3.用交叉熵進行多分類問題

4.用nn.Linear構建全連接層

5. 繼承nn.Module構建全連接層

週末總結——7月5號

交叉熵+全連接層的構建

2D函數最小值優化

visdom——數據可視化

週末總結——6月14日

https://yachay.unat.edu.pe/blog/index.php?comment_area=format_blog&comment_component=blog&comment_co

linux以太網驅動總結