【pytorch】--基于CIFAR10的一个深度学习网络的例子

最新推荐文章于 2024-06-01 23:35:30 发布

YoonJun

最新推荐文章于 2024-06-01 23:35:30 发布

阅读量430

点赞数 8

文章标签： python pytorch

本文链接：https://blog.csdn.net/Yjunshan/article/details/135123169

版权

这篇笔记主要记录一下如何利用torch基于CIFAR10数据设计一个简单的深度学习网络实验

1. 导入第三方库

import torch
import torchvision
from tensorboardX import SummaryWriter
from torch import nn
from torch.nn import Sequential, Conv2d, MaxPool2d, Flatten, Linear
from torch.utils.data import DataLoader

2. 加载数据

train_data = torchvision.datasets.CIFAR10("./CIFAR10_data", train=True, download=True)

Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./CIFAR10_data\cifar-10-python.tar.gz
Failed download. Trying https -> http instead. Downloading http://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./CIFAR10_data\cifar-10-python.tar.gz


100.0%


Extracting ./CIFAR10_data\cifar-10-python.tar.gz to ./CIFAR10_data

# 查看数据类型----> PIL ，但是torch训练网络需要tensor类型的数据.
# 只需要在加载数据的时候经行设置即可---->transform=torchvision.transforms.ToTensor()
train_data[1]

(<PIL.Image.Image image mode=RGB size=32x32>, 9)

train_data = torchvision.datasets.CIFAR10("./CIFAR10_data", train=True, download=True, transform=torchvision.transforms.ToTensor())

Files already downloaded and verified

# 再次查看
train_data[1]

(tensor([[[0.6039, 0.4941, 0.4118,  ..., 0.3569, 0.3412, 0.3098],
          [0.5490, 0.5686, 0.4902,  ..., 0.3765, 0.3020, 0.2784],
          [0.5490, 0.5451, 0.4510,  ..., 0.3098, 0.2667, 0.2627],
          ...,
          [0.6863, 0.6118, 0.6039,  ..., 0.1647, 0.2392, 0.3647],
          [0.6471, 0.6118, 0.6235,  ..., 0.4039, 0.4824, 0.5137],
          [0.6392, 0.6196, 0.6392,  ..., 0.5608, 0.5608, 0.5608]],
 
         [[0.6941, 0.5373, 0.4078,  ..., 0.3725, 0.3529, 0.3176],
          [0.6275, 0.6000, 0.4902,  ..., 0.3882, 0.3137, 0.2863],
          [0.6078, 0.5725, 0.4510,  ..., 0.3216, 0.2745, 0.2706],
          ...,
          [0.6549, 0.6039, 0.6275,  ..., 0.1333, 0.2078, 0.3255],
          [0.6039, 0.5961, 0.6314,  ..., 0.3647, 0.4471, 0.4745],
          [0.5804, 0.5804, 0.6118,  ..., 0.5216, 0.5255, 0.5216]],
 
         [[0.7333, 0.5333, 0.3725,  ..., 0.2784, 0.2784, 0.2745],
          [0.6627, 0.6039, 0.4627,  ..., 0.3059, 0.2431, 0.2392],
          [0.6431, 0.5843, 0.4392,  ..., 0.2510, 0.2157, 0.2157],
          ...,
          [0.6510, 0.6275, 0.6667,  ..., 0.1412, 0.2235, 0.3569],
          [0.5020, 0.5098, 0.5569,  ..., 0.3765, 0.4706, 0.5137],
          [0.4706, 0.4784, 0.5216,  ..., 0.5451, 0.5569, 0.5647]]]),
 9)

test_data = torchvision.datasets.CIFAR10("./CIFAR10_data", train=False, download=True, transform=torchvision.transforms.ToTensor())

Files already downloaded and verified

# 计算训练集以及测试集的大小
train_data_len = len(train_data)
test_data_len = len(test_data)
print('训练集的长度：{}'.format(train_data_len))
print('训练集的长度：{}'.format(test_data_len))

训练集的长度：50000
训练集的长度：10000

# 加载数据
train_data_loader = DataLoader(train_data, batch_size=64)
test_data_loader = DataLoader(test_data, batch_size=64)
print(train_data_loader)
print(test_data_loader)

<torch.utils.data.dataloader.DataLoader object at 0x000002303A7E83D0>
<torch.utils.data.dataloader.DataLoader object at 0x000002303A7E8070>

3. 建神经网络

class Test(nn.Module):
    def __init__(self):
        super().__init__()
        # CIFAR10 model
        self.model1 = Sequential(
            Conv2d(3, 32, 5, padding=2),
            MaxPool2d(2),
            Conv2d(32, 32, 5, padding=2),
            MaxPool2d(2),
            Conv2d(32, 64, 5, padding=2),
            MaxPool2d(2),
            Flatten(),
            Linear(64*4*4, 64),
            Linear(64, 10)
        )
    def forward(self, x):
        x = self.model1(x)
        return x

3.1 测试网络

test = Test()
input_x = torch.ones((64, 3, 32, 32))
out_y = test(input_x)
out_y.shape

torch.Size([64, 10])

out_y[1]

tensor([ 0.0572,  0.0505,  0.0437,  0.2060,  0.0308, -0.0981,  0.0721,  0.1023,
         0.1390,  0.0337], grad_fn=<SelectBackward0>)

3.2 设置损失函数

loss_fun = nn.CrossEntropyLoss()

3.3 设置优化函数

learning_rate = 1e-2
optimizer = torch.optim.SGD(test.parameters(), lr=learning_rate,)

3.4 设置训练网络的一些参数

## 记录训练的次数
total_train_step = 0
## 记录测试的次数
total_test_step = 0
## 记录训练的轮数
total_epoch = 10

## 设置tensorboard
writer = SummaryWriter("./YonJun_log")

for i in range(total_epoch):
    print("------第 {} 轮训练开始-------".format(i+1))
    # 加载训练数据
    for data in train_data_loader:
        imgs, target = data
        output_y = test(imgs)
        loss = loss_fun(output_y, target)
        # print(loss)

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

        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0: # 每训练100次打印一次结果(避免窗口输出太多了信息了)
            print("training:{}, Loss:{}".format(total_train_step, loss.item()))
            writer.add_scalar("train_loss", loss.item(),total_train_step)
    # 测试步骤
    total_test_loss = 0
    with torch.no_grad():
        for data in test_data_loader:
            imgs, target = data
            output_y = test(imgs)
            loss = loss_fun(output_y, target)
            total_test_loss = total_test_loss + loss.item()
    print("Loss:{}".format(total_test_loss))
    writer.add_scalar("test_loss", total_test_loss, total_test_step)
    total_test_step = total_test_step + 1

    # 保存模型
    torch.save(Test, "model_{}.pth".format(i))
writer.close()

------第 1 轮训练开始-------
training:100, Loss:0.7551549673080444
training:200, Loss:0.8748487234115601
training:300, Loss:0.9187159538269043
training:400, Loss:0.8521456122398376
training:500, Loss:0.8002258539199829
training:600, Loss:0.9816480278968811
training:700, Loss:1.0484687089920044
Loss:178.13174575567245
------第 2 轮训练开始-------
training:800, Loss:0.9281576871871948
training:900, Loss:0.8070344924926758
training:1000, Loss:1.0724843740463257
training:1100, Loss:0.9638757705688477
training:1200, Loss:0.8384309411048889
training:1300, Loss:0.804553747177124
training:1400, Loss:0.7303808927536011
training:1500, Loss:0.8349727988243103
Loss:177.53056985139847
------第 3 轮训练开始-------
training:1600, Loss:0.7728996872901917
training:1700, Loss:0.7708233594894409
training:1800, Loss:0.8114984631538391
training:1900, Loss:0.9432680010795593
training:2000, Loss:1.070975422859192
training:2100, Loss:0.6631872057914734
training:2200, Loss:0.6893828511238098
training:2300, Loss:1.023859977722168
Loss:177.24110907316208
------第 4 轮训练开始-------
training:2400, Loss:0.8516899943351746
training:2500, Loss:0.7806205153465271
training:2600, Loss:0.9023863077163696
training:2700, Loss:0.8630157709121704
training:2800, Loss:0.7512081265449524
training:2900, Loss:1.0178464651107788
training:3000, Loss:0.7599277496337891
training:3100, Loss:0.950702965259552
Loss:176.2780049443245
------第 5 轮训练开始-------
training:3200, Loss:0.6675383448600769
training:3300, Loss:0.8793559074401855
training:3400, Loss:0.8816739320755005
training:3500, Loss:0.8316246867179871
training:3600, Loss:0.8430349826812744
training:3700, Loss:0.7882995009422302
training:3800, Loss:0.9813624024391174
training:3900, Loss:0.9530025124549866
Loss:175.99834841489792
------第 6 轮训练开始-------
training:4000, Loss:0.8220319747924805
training:4100, Loss:0.8931095004081726
training:4200, Loss:1.0225799083709717
training:4300, Loss:0.8239660859107971
training:4400, Loss:0.5933955311775208
training:4500, Loss:0.9415645003318787
training:4600, Loss:0.9078661203384399
Loss:175.51610720157623
------第 7 轮训练开始-------
training:4700, Loss:0.670432984828949
training:4800, Loss:0.8908045887947083
training:4900, Loss:0.8217464685440063
training:5000, Loss:0.8287463188171387
training:5100, Loss:0.5940623879432678
training:5200, Loss:0.8417458534240723
training:5300, Loss:0.6588895916938782
training:5400, Loss:0.6623363494873047
Loss:175.0481561422348
------第 8 轮训练开始-------
training:5500, Loss:0.7641991972923279
training:5600, Loss:0.6625202894210815
training:5700, Loss:0.5682169795036316
training:5800, Loss:0.6444393992424011
training:5900, Loss:0.8577033877372742
training:6000, Loss:0.9969536066055298
training:6100, Loss:0.6678980588912964
training:6200, Loss:0.7402710318565369
Loss:174.88561528921127
------第 9 轮训练开始-------
training:6300, Loss:0.7923698425292969
training:6400, Loss:0.6133146286010742
training:6500, Loss:0.9324386715888977
training:6600, Loss:0.6652389168739319
training:6700, Loss:0.599393904209137
training:6800, Loss:0.6970618963241577
training:6900, Loss:0.6301713585853577
training:7000, Loss:0.37432461977005005
Loss:173.943276822567
------第 10 轮训练开始-------
training:7100, Loss:0.6298545598983765
training:7200, Loss:0.5420732498168945
training:7300, Loss:0.6390514373779297
training:7400, Loss:0.48693612217903137
training:7500, Loss:0.6869808435440063
training:7600, Loss:0.9039825201034546
training:7700, Loss:0.5407873392105103
training:7800, Loss:0.8126641511917114
Loss:177.13708823919296