天天看點

mseloss pytorch_60題PyTorch簡易入門指南,做技術的弄潮兒

PyTorch是一個基于Python的庫,提供了一個具有靈活易用的深度學習架構,是近年來最受歡迎的深度學習架構之一。

點選此處檢視線上運作效果

和鲸社群 - Kesci.com​www.kesci.com

如果你是新新新手,可以先學習以下教程:

  • 深度學習之PyTorch實戰-基礎學習及搭建環境
  • PyTorch中文文檔

改編自:

  • DEEP LEARNING WITH PYTORCH: A 60 MINUTE BLITZ

其他x題系列:

  • 35題初探scikit-learn庫,get機器學習好幫手√
  • 50題matplotlib從入門到精通
  • 50道練習帶你玩轉Pandas
  • 一兩贅肉無:100道練習帶你玩轉Numpy

1 初識PyTorch

1.1 張量

1.導入pytorch包

import torch
           

2.建立一個空的5x3張量

x = torch.empty(5, 3)
print(x)
           

3.建立一個随機初始化的5x3張量

x = torch.rand(5, 3)
print(x)
           

4.建立一個5x3的0張量,類型為long

x = torch.zeros(5, 3, dtype=torch.long)
print(x)
           

5.直接從數組建立張量

x = torch.tensor([5.5, 3])
print(x)
           

6.建立一個5x3的機關張量,類型為double

x = torch.ones(5, 3, dtype=torch.double)
print(x)
           

7.從已有的張量建立相同次元的新張量,并且重新定義類型為float

x = torch.randn_like(x, dtype=torch.float)
print(x)
           

8.列印一個張量的次元

print(x.size())
           

9.将兩個張量相加

y = torch.rand(5, 3)
print(x + y)

# 方法二
# print(torch.add(x, y))

# 方法三
# result = torch.empty(5, 3)
# torch.add(x, y, out=result)
# print(result)

# 方法四
# y.add_(x)
# print(y)
           

10.取張量的第一列

print(x[:, 1])
           

11.将一個4x4的張量resize成一個一維張量

x = torch.randn(4, 4)
y = x.view(16)
print(x.size(),y.size())
           

12.将一個4x4的張量,resize成一個2x8的張量

y = x.view(2, 8)
print(x.size(),y.size())

# 方法二
z = x.view(-1, 8) # 确定一個次元,-1的次元會被自動計算
print(x.size(),z.size())
           

13.從張量中取出數字

x = torch.randn(1)
print(x)
print(x.item())
           

1.2 Numpy的操作

14.将張量裝換成numpy數組

a = torch.ones(5)
print(a)

b = a.numpy()
print(b)
           

15.将張量+1,并觀察上題中numpy數組的變化

a.add_(1)
print(a)
print(b)
           

16.從numpy數組建立張量

import numpy as np
a = np.ones(5)
b = torch.from_numpy(a)
print(a)
print(b)
           

17.将numpy數組+1并觀察上題中張量的變化

np.add(a, 1, out=a)
print(a)
print(b)
           

2 自動微分

2.1 張量的自動微分

18.建立一個張量,并設定

requires_grad=True

x = torch.ones(2, 2, requires_grad=True)
print(x)
           

19.對張量進行任意操作(y = x + 2)

y = x + 2
print(y)
print(y.grad_fn) # y就多了一個AddBackward
           

20.再對y進行任意操作

z = y * y * 3
out = z.mean()

print(z) # z多了MulBackward
print(out) # out多了MeanBackward
           

2.2 梯度

21.對out進行反向傳播

out.backward()
           

22.列印梯度d(out)/dx

print(x.grad) #out=0.25*Σ3(x+2)^2
           

23.建立一個結果為矢量的計算過程(y=x*2^n)

x = torch.randn(3, requires_grad=True)

y = x * 2
while y.data.norm() < 1000:
    y = y * 2

print(y)
           

24.計算

v = [0.1, 1.0, 0.0001]

處的梯度

v = torch.tensor([0.1, 1.0, 0.0001], dtype=torch.float)
y.backward(v)

print(x.grad)
           

25.關閉梯度的功能

print(x.requires_grad)
print((x ** 2).requires_grad)

with torch.no_grad():
    print((x ** 2).requires_grad)
    
# 方法二
# print(x.requires_grad)
# y = x.detach()
# print(y.requires_grad)
# print(x.eq(y).all())
           

3 神經網絡

這部分會實作LeNet5,結構如下所示

mseloss pytorch_60題PyTorch簡易入門指南,做技術的弄潮兒

3.1 定義網絡

import torch
import torch.nn as nn
import torch.nn.functional as F


class Net(nn.Module):

    def __init__(self):
        super(Net, self).__init__()
        # 26.定義①的卷積層,輸入為32x32的圖像,卷積核大小5x5卷積核種類6
        self.conv1 = nn.Conv2d(3, 6, 5)
        # 27.定義③的卷積層,輸入為前一層6個特征,卷積核大小5x5,卷積核種類16
        self.conv2 = nn.Conv2d(6, 16, 5)
        # 28.定義⑤的全連結層,輸入為16*5*5,輸出為120
        self.fc1 = nn.Linear(16 * 5 * 5, 120)  # 6*6 from image dimension
        # 29.定義⑥的全連接配接層,輸入為120,輸出為84
        self.fc2 = nn.Linear(120, 84)
        # 30.定義⑥的全連接配接層,輸入為84,輸出為10
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        # 31.完成input-S2,先卷積+relu,再2x2下采樣
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        # 32.完成S2-S4,先卷積+relu,再2x2下采樣
        x = F.max_pool2d(F.relu(self.conv2(x)), 2) #卷積核方形時,可以隻寫一個次元
        # 33.将特征向量扁平成行向量
        x = x.view(-1, 16 * 5 * 5)
        # 34.使用fc1+relu
        x = F.relu(self.fc1(x))
        # 35.使用fc2+relu
        x = F.relu(self.fc2(x))
        # 36.使用fc3
        x = self.fc3(x)
        return x


net = Net()
print(net)
           

37.列印網絡的參數

params = list(net.parameters())
# print(params)
print(len(params))
           

38.列印某一層參數的形狀

print(params[0].size())
           

39.随機輸入一個向量,檢視前向傳播輸出

input = torch.randn(1, 1, 32, 32)
out = net(input)
print(out)
           

40.将梯度初始化

net.zero_grad()
           

41.随機一個梯度進行反向傳播

out.backward(torch.randn(1, 10))
           

3.2 損失函數

42.用自帶的MSELoss()定義損失函數

criterion = nn.MSELoss()
           

43.随機一個真值,并用随機的輸入計算損失

target = torch.randn(10)  # 随機真值
target = target.view(1, -1)  # 變成行向量

output = net(input)  # 用随機輸入計算輸出

loss = criterion(output, target)  # 計算損失
print(loss)
           

44.将梯度初始化,計算上一步中loss的反向傳播

net.zero_grad()

print('conv1.bias.grad before backward')
print(net.conv1.bias.grad)
           

45.計算43中loss的反向傳播

loss.backward()

print('conv1.bias.grad after backward')
print(net.conv1.bias.grad)
           

3.3 更新權重

46.定義SGD優化器算法,學習率設定為0.01

import torch.optim as optim
optimizer = optim.SGD(net.parameters(), lr=0.01)
           

47.使用優化器更新權重

optimizer.zero_grad()
output = net(input)
loss = criterion(output, target)
loss.backward()

# 更新權重
optimizer.step()
           

4 訓練一個分類器

4.1 讀取CIFAR10資料,做标準化

48.構造一個transform,将三通道(0,1)區間的資料轉換成(-1,1)的資料

import torchvision
import torchvision.transforms as transforms

transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
           

讀取資料集

trainset = cifar(root = './input/cifar10', segmentation='train', transforms=transform)
testset = cifar(root = './input/cifar10', segmentation='test', transforms=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size,shuffle=True, num_workers=2)
testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size,shuffle=False, num_workers=2) 

classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
           

4.2 建立網絡

這部分沿用前面的網絡

net2 = Net()
           

4.3 定義損失函數和優化器

49.定義交叉熵損失函數

criterion2 = nn.CrossEntropyLoss()
           

50.定義SGD優化器算法,學習率設定為0.001,

momentum=0.9

optimizer2 = optim.SGD(net2.parameters(), lr=0.001, momentum=0.9)
           

4.4訓練網絡

for epoch in range(2): 

    running_loss = 0.0
    for i, data in enumerate(trainloader, 0):
        # 擷取X,y對
        inputs, labels = data

        # 51.初始化梯度
        optimizer2.zero_grad()

        # 52.前饋
        outputs = net2(inputs)
        # 53.計算損失
        loss = criterion2(outputs, labels)
        # 54.計算梯度
        loss.backward()
        # 55.更新權值
        optimizer2.step()

        # 每2000個資料列印平均代價函數值
        running_loss += loss.item()
        if i % 2000 == 1999:    # print every 2000 mini-batches
            print('[%d, %5d] loss: %.3f' %
                  (epoch + 1, i + 1, running_loss / 2000))
            running_loss = 0.0

print('Finished Training')
           

4.5 使用模型預測

取一些資料

dataiter = iter(testloader)
images, labels = dataiter.next()

# print images
imshow(torchvision.utils.make_grid(images))
print('GroundTruth: ', ' '.join('%5s' % classes[labels[j]] for j in range(4)))
           

56.使用模型預測

outputs = net2(images)

_, predicted = torch.max(outputs, 1)

print('Predicted: ', ' '.join('%5s' % classes[predicted[j]]
                              for j in range(4)))
           

57.在測試集上進行打分

correct = 0
total = 0
with torch.no_grad():
    for data in testloader:
        images, labels = data
        outputs = net2(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

print('Accuracy of the network on the 10000 test images: %d %%' % (
    100 * correct / total))
           

4.6 存取模型

58.儲存訓練好的模型

PATH = './cifar_net.pth'
torch.save(net.state_dict(), PATH)
           

59.讀取儲存的模型

pretrained_net = torch.load(PATH)
           

60.加載模型

net3 = Net()

net3.load_state_dict(pretrained_net)
           

繼續閱讀