文章目录
- 1.简介
- 2.应用
- 3.演进
- 4.机器学习
- 5.深度学习
- 6.强化学习
- 7.图像识别
- 8.自然语言
- 9.Python
- 10.Python开发环境
- 11.机器学习算法
- 12.神经网络
- 13.PyTorch实战，手写数字识别
1.简介
2.应用
3.演进
4.机器学习
5.深度学习
6.强化学习
7.图像识别
8.自然语言
9.Python
10.Python开发环境
11.机器学习算法
1.多元线性回归
项自实战：糖尿病回归预测
配置清华镜像
2.逻辑回归
3.Softmax回归
项目实战：鸢尾花大作战
4.正则化技术
项目实战：新闻分类
解决fetch_20newsgroups数据集无法加载403问题
5.梯度下降法
6.数据归一化
项目实战：手写数字识别
7.KMeans聚类
项目实战：KMeans聚类代码实现
8.高斯混合模型
项目实战：说话人识别
12.神经网络
1.感知机
2.神经网络
3.激活函数
4.正向反向传播
5.梯度消失
6.Dropout
13.PyTorch实战，手写数字识别
python
import torch
from torchvision import datasets, transforms
# print(torch.__version__)
#检测CUDA是否可用
use_cuda = torch.cuda.is_available()
# print(use_cuda)
# 设置device变量并
if use_cuda:
device = torch.device("cuda")
else:
device = torch.device("cpu")
transform = transforms.Compose([
#让数据转成Tensor张量
transforms.ToTensor()
# 让图片数据进行标准归一化，0.1307是标准归一化的均值，0.3081对应的是标准归一化的方差
#transforms.Normalize((0.1307)，(.3081,))
])
# 读取数据
datasets1 = datasets.MNIST('./data', train=True, download=True, transform=transform)
datasets2 = datasets.MNIST('./data', train=False, download=True, transform=transform)
#设置数据加载器，顺带手设置批次大小和是否打乱数据顺序
train_loader = torch.utils.data.DataLoader(datasets1, batch_size=60000, shuffle=True)
test_loader = torch.utils.data.DataLoader(datasets2, batch_size=1000)
for batch_idx, data in enumerate(train_loader, 0):
inputs, targets = data
# view在下一行会把我们的训练集(60000,1,28,28)转换成(60000,28*28)
x = inputs.view(-1, 28 * 28)
#计算所有训练样本的标准差和均值
x_std = x.std().item()
x_mean = x.mean().item()
print('均值mean为:' + str(x_mean))
print('标准差std为:' + str(x_std))
python
import torch
from torchvision import datasets, transforms
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
# print(torch.__version__)
#检测CUDA是否可用
use_cuda = torch.cuda.is_available()
# print(use_cuda)
# 设置device变量并
if use_cuda:
device = torch.device("cuda")
else:
device = torch.device("cpu")
transform = transforms.Compose([
#让数据转成Tensor张量
transforms.ToTensor(),
# 让图片数据进行标准归一化，0.1307是标准归一化的均值，0.3081对应的是标准归一化的方差
transforms.Normalize((0.1307,), (0.3081,))
])
# 读取数据
datasets1 = datasets.MNIST('./data', train=True, download=True, transform=transform)
datasets2 = datasets.MNIST('./data', train=False, download=True, transform=transform)
#设置数据加载器，顺带手设置批次大小和是否打乱数据顺序
train_loader = torch.utils.data.DataLoader(datasets1, batch_size=128, shuffle=True)
test_loader = torch.utils.data.DataLoader(datasets2, batch_size=1000)
# for batch_idx, data in enumerate(train_loader, 0):
# inputs, targets = data
# # view在下一行会把我们的训练集(60000,1,28,28)转换成(60000,28*28)
# x = inputs.view(-1, 28 * 28)
# #计算所有训练样本的标准差和均值
# x_std = x.std().item()
# x_mean = x.mean().item()
# print('均值mean为:' + str(x_mean))
# print('标准差std为:' + str(x_std))
#通过自定义类来构建模型
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.fc1 = nn.Linear(784, 128)
self.dropout = nn.Dropout(0.2)
self.fc2 = nn.Linear(128, 10)
def forward(self, x):
x = torch.flatten(x, 1)
x = self.fc1(x)
x = F.relu(x)
x = self.dropout(x)
x = self.fc2(x)
output = F.log_softmax(x, dim=1)
return output
#创建一个模型实例
model = Net().to(device)
#定义训练模型的逻辑
def train_step(data, target, model, optimizer):
optimizer.zero_grad()
output = model(data)
#nll代表着negative log likely hood 负对数似然
loss = F.nll_loss(output, target)
#反向传播的本质是不是就是去求梯度
loss.backward()
#本质就是应用梯度去调参
optimizer.step()
return loss
#定义测试模型的逻辑
def test_step(data, target, model, test_loss, correct):
output = model(data)
# 累积的批次损失
test_loss += F.nll_loss(output, target, reduction='sum').item()
#获得对数概率最大值对应的索引号，这里其实就是类别号
pred = output.argmax(dim=1, keepdims=True)
correct += pred.eq(target.view_as(pred)).sum().item()
return test_loss, correct
# 创建训练调参使用的优化器
optimizer = optim.Adam(model.parameters(), lr=0.001)
#真正的分轮次训练
EPOCHS = 5
for epoch in range(EPOCHS):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
loss = train_step(data, target, model, optimizer)
# 每隔10个批次，打印信息
if batch_idx % 10 == 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()))
model.eval()
test_loss = 0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.to(device), target.to(device)
test_loss, correct = test_step(data, target, model, test_loss, correct)
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)))
bash
D:\ProgramData\miniconda3\envs\pytorch113\python.exe "D:\ProgramData\AIProject\AI\PyTorch_Study\mnist _dnn.py"
Train Epoch: 0 [0/60000 (0%)] Loss:2.357153
Train Epoch: 0 [1280/60000 (2%)] Loss:1.188171
Train Epoch: 0 [2560/60000 (4%)] Loss:0.778216
Train Epoch: 0 [3840/60000 (6%)] Loss:0.527082
Train Epoch: 0 [5120/60000 (9%)] Loss:0.449032
Train Epoch: 0 [6400/60000 (11%)] Loss:0.439059
Train Epoch: 0 [7680/60000 (13%)] Loss:0.456310
Train Epoch: 0 [8960/60000 (15%)] Loss:0.340990
Train Epoch: 0 [10240/60000 (17%)] Loss:0.537856
Train Epoch: 0 [11520/60000 (19%)] Loss:0.359342
Train Epoch: 0 [12800/60000 (21%)] Loss:0.373274
Train Epoch: 0 [14080/60000 (23%)] Loss:0.251031
Train Epoch: 0 [15360/60000 (26%)] Loss:0.360636
Train Epoch: 0 [16640/60000 (28%)] Loss:0.302707
Train Epoch: 0 [17920/60000 (30%)] Loss:0.185309
Train Epoch: 0 [19200/60000 (32%)] Loss:0.299687
Train Epoch: 0 [20480/60000 (34%)] Loss:0.435019
Train Epoch: 0 [21760/60000 (36%)] Loss:0.208732
Train Epoch: 0 [23040/60000 (38%)] Loss:0.335426
Train Epoch: 0 [24320/60000 (41%)] Loss:0.301936
Train Epoch: 0 [25600/60000 (43%)] Loss:0.237923
Train Epoch: 0 [26880/60000 (45%)] Loss:0.243458
Train Epoch: 0 [28160/60000 (47%)] Loss:0.258387
Train Epoch: 0 [29440/60000 (49%)] Loss:0.324313
Train Epoch: 0 [30720/60000 (51%)] Loss:0.226612
Train Epoch: 0 [32000/60000 (53%)] Loss:0.286255
Train Epoch: 0 [33280/60000 (55%)] Loss:0.286046
Train Epoch: 0 [34560/60000 (58%)] Loss:0.319120
Train Epoch: 0 [35840/60000 (60%)] Loss:0.235170
Train Epoch: 0 [37120/60000 (62%)] Loss:0.234801
Train Epoch: 0 [38400/60000 (64%)] Loss:0.172543
Train Epoch: 0 [39680/60000 (66%)] Loss:0.171685
Train Epoch: 0 [40960/60000 (68%)] Loss:0.223411
Train Epoch: 0 [42240/60000 (70%)] Loss:0.181646
Train Epoch: 0 [43520/60000 (72%)] Loss:0.236268
Train Epoch: 0 [44800/60000 (75%)] Loss:0.147353
Train Epoch: 0 [46080/60000 (77%)] Loss:0.404344
Train Epoch: 0 [47360/60000 (79%)] Loss:0.210359
Train Epoch: 0 [48640/60000 (81%)] Loss:0.193106
Train Epoch: 0 [49920/60000 (83%)] Loss:0.213325
Train Epoch: 0 [51200/60000 (85%)] Loss:0.239207
Train Epoch: 0 [52480/60000 (87%)] Loss:0.194574
Train Epoch: 0 [53760/60000 (90%)] Loss:0.130250
Train Epoch: 0 [55040/60000 (92%)] Loss:0.174132
Train Epoch: 0 [56320/60000 (94%)] Loss:0.157513
Train Epoch: 0 [57600/60000 (96%)] Loss:0.210445
Train Epoch: 0 [58880/60000 (98%)] Loss:0.178082
Test set: Average loss: 0.1582, Accuracy: 9543/10000 (95%)
Train Epoch: 1 [0/60000 (0%)] Loss:0.190592
Train Epoch: 1 [1280/60000 (2%)] Loss:0.137684
Train Epoch: 1 [2560/60000 (4%)] Loss:0.271779
Train Epoch: 1 [3840/60000 (6%)] Loss:0.212214
Train Epoch: 1 [5120/60000 (9%)] Loss:0.293748
Train Epoch: 1 [6400/60000 (11%)] Loss:0.230604
Train Epoch: 1 [7680/60000 (13%)] Loss:0.254167
Train Epoch: 1 [8960/60000 (15%)] Loss:0.168410
Train Epoch: 1 [10240/60000 (17%)] Loss:0.146805
Train Epoch: 1 [11520/60000 (19%)] Loss:0.190707
Train Epoch: 1 [12800/60000 (21%)] Loss:0.173118
Train Epoch: 1 [14080/60000 (23%)] Loss:0.129635
Train Epoch: 1 [15360/60000 (26%)] Loss:0.176186
Train Epoch: 1 [16640/60000 (28%)] Loss:0.122384
Train Epoch: 1 [17920/60000 (30%)] Loss:0.163052
Train Epoch: 1 [19200/60000 (32%)] Loss:0.143208
Train Epoch: 1 [20480/60000 (34%)] Loss:0.249731
Train Epoch: 1 [21760/60000 (36%)] Loss:0.141281
Train Epoch: 1 [23040/60000 (38%)] Loss:0.245926
Train Epoch: 1 [24320/60000 (41%)] Loss:0.224928
Train Epoch: 1 [25600/60000 (43%)] Loss:0.226181
Train Epoch: 1 [26880/60000 (45%)] Loss:0.098913
Train Epoch: 1 [28160/60000 (47%)] Loss:0.138237
Train Epoch: 1 [29440/60000 (49%)] Loss:0.145725
Train Epoch: 1 [30720/60000 (51%)] Loss:0.142157
Train Epoch: 1 [32000/60000 (53%)] Loss:0.085880
Train Epoch: 1 [33280/60000 (55%)] Loss:0.160553
Train Epoch: 1 [34560/60000 (58%)] Loss:0.133455
Train Epoch: 1 [35840/60000 (60%)] Loss:0.129697
Train Epoch: 1 [37120/60000 (62%)] Loss:0.241462
Train Epoch: 1 [38400/60000 (64%)] Loss:0.138529
Train Epoch: 1 [39680/60000 (66%)] Loss:0.147758
Train Epoch: 1 [40960/60000 (68%)] Loss:0.223078
Train Epoch: 1 [42240/60000 (70%)] Loss:0.136696
Train Epoch: 1 [43520/60000 (72%)] Loss:0.162502
Train Epoch: 1 [44800/60000 (75%)] Loss:0.201067
Train Epoch: 1 [46080/60000 (77%)] Loss:0.118544
Train Epoch: 1 [47360/60000 (79%)] Loss:0.075555
Train Epoch: 1 [48640/60000 (81%)] Loss:0.187998
Train Epoch: 1 [49920/60000 (83%)] Loss:0.158163
Train Epoch: 1 [51200/60000 (85%)] Loss:0.117933
Train Epoch: 1 [52480/60000 (87%)] Loss:0.074444
Train Epoch: 1 [53760/60000 (90%)] Loss:0.076563
Train Epoch: 1 [55040/60000 (92%)] Loss:0.126042
Train Epoch: 1 [56320/60000 (94%)] Loss:0.104785
Train Epoch: 1 [57600/60000 (96%)] Loss:0.124018
Train Epoch: 1 [58880/60000 (98%)] Loss:0.131464
Test set: Average loss: 0.1128, Accuracy: 9647/10000 (96%)
Train Epoch: 2 [0/60000 (0%)] Loss:0.135893
Train Epoch: 2 [1280/60000 (2%)] Loss:0.107287
Train Epoch: 2 [2560/60000 (4%)] Loss:0.092717
Train Epoch: 2 [3840/60000 (6%)] Loss:0.162008
Train Epoch: 2 [5120/60000 (9%)] Loss:0.118251
Train Epoch: 2 [6400/60000 (11%)] Loss:0.167732
Train Epoch: 2 [7680/60000 (13%)] Loss:0.099356
Train Epoch: 2 [8960/60000 (15%)] Loss:0.058201
Train Epoch: 2 [10240/60000 (17%)] Loss:0.087654
Train Epoch: 2 [11520/60000 (19%)] Loss:0.103737
Train Epoch: 2 [12800/60000 (21%)] Loss:0.092910
Train Epoch: 2 [14080/60000 (23%)] Loss:0.168416
Train Epoch: 2 [15360/60000 (26%)] Loss:0.120313
Train Epoch: 2 [16640/60000 (28%)] Loss:0.179698
Train Epoch: 2 [17920/60000 (30%)] Loss:0.195164
Train Epoch: 2 [19200/60000 (32%)] Loss:0.100405
Train Epoch: 2 [20480/60000 (34%)] Loss:0.127287
Train Epoch: 2 [21760/60000 (36%)] Loss:0.096777
Train Epoch: 2 [23040/60000 (38%)] Loss:0.128975
Train Epoch: 2 [24320/60000 (41%)] Loss:0.174980
Train Epoch: 2 [25600/60000 (43%)] Loss:0.153128
Train Epoch: 2 [26880/60000 (45%)] Loss:0.202121
Train Epoch: 2 [28160/60000 (47%)] Loss:0.098918
Train Epoch: 2 [29440/60000 (49%)] Loss:0.106479
Train Epoch: 2 [30720/60000 (51%)] Loss:0.094558
Train Epoch: 2 [32000/60000 (53%)] Loss:0.104092
Train Epoch: 2 [33280/60000 (55%)] Loss:0.075859
Train Epoch: 2 [34560/60000 (58%)] Loss:0.155020
Train Epoch: 2 [35840/60000 (60%)] Loss:0.172556
Train Epoch: 2 [37120/60000 (62%)] Loss:0.054027
Train Epoch: 2 [38400/60000 (64%)] Loss:0.242806
Train Epoch: 2 [39680/60000 (66%)] Loss:0.150216
Train Epoch: 2 [40960/60000 (68%)] Loss:0.131288
Train Epoch: 2 [42240/60000 (70%)] Loss:0.055607
Train Epoch: 2 [43520/60000 (72%)] Loss:0.137465
Train Epoch: 2 [44800/60000 (75%)] Loss:0.141450
Train Epoch: 2 [46080/60000 (77%)] Loss:0.120138
Train Epoch: 2 [47360/60000 (79%)] Loss:0.086141
Train Epoch: 2 [48640/60000 (81%)] Loss:0.149587
Train Epoch: 2 [49920/60000 (83%)] Loss:0.074414
Train Epoch: 2 [51200/60000 (85%)] Loss:0.106521
Train Epoch: 2 [52480/60000 (87%)] Loss:0.082931
Train Epoch: 2 [53760/60000 (90%)] Loss:0.085414
Train Epoch: 2 [55040/60000 (92%)] Loss:0.166222
Train Epoch: 2 [56320/60000 (94%)] Loss:0.164097
Train Epoch: 2 [57600/60000 (96%)] Loss:0.115938
Train Epoch: 2 [58880/60000 (98%)] Loss:0.144959
Test set: Average loss: 0.0957, Accuracy: 9718/10000 (97%)
Train Epoch: 3 [0/60000 (0%)] Loss:0.145348
Train Epoch: 3 [1280/60000 (2%)] Loss:0.129786
Train Epoch: 3 [2560/60000 (4%)] Loss:0.068685
Train Epoch: 3 [3840/60000 (6%)] Loss:0.044195
Train Epoch: 3 [5120/60000 (9%)] Loss:0.077902
Train Epoch: 3 [6400/60000 (11%)] Loss:0.108578
Train Epoch: 3 [7680/60000 (13%)] Loss:0.149138
Train Epoch: 3 [8960/60000 (15%)] Loss:0.099387
Train Epoch: 3 [10240/60000 (17%)] Loss:0.103183
Train Epoch: 3 [11520/60000 (19%)] Loss:0.100638
Train Epoch: 3 [12800/60000 (21%)] Loss:0.092041
Train Epoch: 3 [14080/60000 (23%)] Loss:0.073178
Train Epoch: 3 [15360/60000 (26%)] Loss:0.207282
Train Epoch: 3 [16640/60000 (28%)] Loss:0.076489
Train Epoch: 3 [17920/60000 (30%)] Loss:0.148314
Train Epoch: 3 [19200/60000 (32%)] Loss:0.039077
Train Epoch: 3 [20480/60000 (34%)] Loss:0.115610
Train Epoch: 3 [21760/60000 (36%)] Loss:0.133416
Train Epoch: 3 [23040/60000 (38%)] Loss:0.084655
Train Epoch: 3 [24320/60000 (41%)] Loss:0.148035
Train Epoch: 3 [25600/60000 (43%)] Loss:0.152145
Train Epoch: 3 [26880/60000 (45%)] Loss:0.071100
Train Epoch: 3 [28160/60000 (47%)] Loss:0.056352
Train Epoch: 3 [29440/60000 (49%)] Loss:0.227481
Train Epoch: 3 [30720/60000 (51%)] Loss:0.138899
Train Epoch: 3 [32000/60000 (53%)] Loss:0.080404
Train Epoch: 3 [33280/60000 (55%)] Loss:0.066160
Train Epoch: 3 [34560/60000 (58%)] Loss:0.080147
Train Epoch: 3 [35840/60000 (60%)] Loss:0.079220
Train Epoch: 3 [37120/60000 (62%)] Loss:0.058759
Train Epoch: 3 [38400/60000 (64%)] Loss:0.073250
Train Epoch: 3 [39680/60000 (66%)] Loss:0.121640
Train Epoch: 3 [40960/60000 (68%)] Loss:0.166685
Train Epoch: 3 [42240/60000 (70%)] Loss:0.047174
Train Epoch: 3 [43520/60000 (72%)] Loss:0.076186
Train Epoch: 3 [44800/60000 (75%)] Loss:0.115248
Train Epoch: 3 [46080/60000 (77%)] Loss:0.138637
Train Epoch: 3 [47360/60000 (79%)] Loss:0.138182
Train Epoch: 3 [48640/60000 (81%)] Loss:0.138327
Train Epoch: 3 [49920/60000 (83%)] Loss:0.027602
Train Epoch: 3 [51200/60000 (85%)] Loss:0.068769
Train Epoch: 3 [52480/60000 (87%)] Loss:0.125922
Train Epoch: 3 [53760/60000 (90%)] Loss:0.079334
Train Epoch: 3 [55040/60000 (92%)] Loss:0.184128
Train Epoch: 3 [56320/60000 (94%)] Loss:0.110396
Train Epoch: 3 [57600/60000 (96%)] Loss:0.115563
Train Epoch: 3 [58880/60000 (98%)] Loss:0.051400
Test set: Average loss: 0.0802, Accuracy: 9759/10000 (98%)
Train Epoch: 4 [0/60000 (0%)] Loss:0.085800
Train Epoch: 4 [1280/60000 (2%)] Loss:0.047675
Train Epoch: 4 [2560/60000 (4%)] Loss:0.029820
Train Epoch: 4 [3840/60000 (6%)] Loss:0.091837
Train Epoch: 4 [5120/60000 (9%)] Loss:0.076924
Train Epoch: 4 [6400/60000 (11%)] Loss:0.077931
Train Epoch: 4 [7680/60000 (13%)] Loss:0.096714
Train Epoch: 4 [8960/60000 (15%)] Loss:0.043513
Train Epoch: 4 [10240/60000 (17%)] Loss:0.165721
Train Epoch: 4 [11520/60000 (19%)] Loss:0.077272
Train Epoch: 4 [12800/60000 (21%)] Loss:0.036313
Train Epoch: 4 [14080/60000 (23%)] Loss:0.108138
Train Epoch: 4 [15360/60000 (26%)] Loss:0.084638
Train Epoch: 4 [16640/60000 (28%)] Loss:0.157234
Train Epoch: 4 [17920/60000 (30%)] Loss:0.094364
Train Epoch: 4 [19200/60000 (32%)] Loss:0.069152
Train Epoch: 4 [20480/60000 (34%)] Loss:0.014761
Train Epoch: 4 [21760/60000 (36%)] Loss:0.046572
Train Epoch: 4 [23040/60000 (38%)] Loss:0.076240
Train Epoch: 4 [24320/60000 (41%)] Loss:0.064022
Train Epoch: 4 [25600/60000 (43%)] Loss:0.051202
Train Epoch: 4 [26880/60000 (45%)] Loss:0.113288
Train Epoch: 4 [28160/60000 (47%)] Loss:0.105636
Train Epoch: 4 [29440/60000 (49%)] Loss:0.103099
Train Epoch: 4 [30720/60000 (51%)] Loss:0.027316
Train Epoch: 4 [32000/60000 (53%)] Loss:0.029250
Train Epoch: 4 [33280/60000 (55%)] Loss:0.087277
Train Epoch: 4 [34560/60000 (58%)] Loss:0.151358
Train Epoch: 4 [35840/60000 (60%)] Loss:0.106468
Train Epoch: 4 [37120/60000 (62%)] Loss:0.052081
Train Epoch: 4 [38400/60000 (64%)] Loss:0.161355
Train Epoch: 4 [39680/60000 (66%)] Loss:0.058824
Train Epoch: 4 [40960/60000 (68%)] Loss:0.115524
Train Epoch: 4 [42240/60000 (70%)] Loss:0.122241
Train Epoch: 4 [43520/60000 (72%)] Loss:0.097723
Train Epoch: 4 [44800/60000 (75%)] Loss:0.022386
Train Epoch: 4 [46080/60000 (77%)] Loss:0.038309
Train Epoch: 4 [47360/60000 (79%)] Loss:0.056229
Train Epoch: 4 [48640/60000 (81%)] Loss:0.053395
Train Epoch: 4 [49920/60000 (83%)] Loss:0.108051
Train Epoch: 4 [51200/60000 (85%)] Loss:0.062542
Train Epoch: 4 [52480/60000 (87%)] Loss:0.020315
Train Epoch: 4 [53760/60000 (90%)] Loss:0.054988
Train Epoch: 4 [55040/60000 (92%)] Loss:0.083139
Train Epoch: 4 [56320/60000 (94%)] Loss:0.068836
Train Epoch: 4 [57600/60000 (96%)] Loss:0.113024
Train Epoch: 4 [58880/60000 (98%)] Loss:0.076040
Test set: Average loss: 0.0813, Accuracy: 9740/10000 (97%)
进程已结束，退出代码为 0