PyTorch-完整的模型训练套路

目录

1. 准备

1.1 使用Cifar10

1.2 item的用法

1.3 model的搭建

1.4 数据集、参数设置以及训练开始

1.5 测试集

1.6 使用tensorboard

1.7 保存每一轮的训练结果

2. 计算整体的正确率

3. 其他

3.1 train与eval

3.2 使用GPU训练

---

1. 准备

1.1 使用Cifar10

1.2 item的用法

import torch

a = torch.tensor(3)
print(a)
print(a.item())

tensor(3)
3

1.3 model的搭建

model.py

import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear

# 搭建神经网络
class MyModule(nn.Module):
    def __init__(self):
        super(MyModule, self).__init__()
        self.model = nn.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(1024, 64),
            Linear(64, 10)
        )

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

# 测试
if __name__ == '__main__':
    myModule = MyModule()
    input = torch.ones((64, 3, 32, 32))
    output = myModule(input)
    print(output.shape)  # torch.Size([64, 10])

1.4 数据集、参数设置以及训练开始

train.py

import torch.optim.optimizer
import torchvision
from model import *

from torch.utils.data import DataLoader

# 1.数据集准备
train_data = torchvision.datasets.CIFAR10('../dataset', train=True, transform=torchvision.transforms.ToTensor(),
                                          download=True)
test_data = torchvision.datasets.CIFAR10('../dataset', train=False, transform=torchvision.transforms.ToTensor(),
                                         download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
# print(train_data_size, test_data_size)  # 50000 10000

# 2.加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 3.创建网络模型
myModule = MyModule()

# 4.损失函数
loss_fn = nn.CrossEntropyLoss()
# 5.优化器
learningRate = 1e-2
optimizer = torch.optim.SGD(myModule.parameters(), lr=learningRate)
# 6.设置训练网络的一些参数
total_train_step = 0  # 记录训练的次数
total_test_step = 0  # 记录测试的次数
epoch = 10  # 记录训练的轮数
for i in range(epoch):
    # 开始训练
    for data in train_dataloader:
        imgs, targets = data
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        print("训练次数: {},loss: {}".format(total_train_step, loss.item()))

Files already downloaded and verified
Files already downloaded and verified
训练次数: 1,loss: 2.291430711746216
训练次数: 2,loss: 2.294950485229492
训练次数: 3,loss: 2.3185925483703613
训练次数: 4,loss: 2.2968363761901855
训练次数: 5,loss: 2.30112886428833
训练次数: 6,loss: 2.3146629333496094
训练次数: 7,loss: 2.3073482513427734
训练次数: 8,loss: 2.3127682209014893

......

1.5 测试集

目的：可以拿测试集来验证模型训练的怎么样了。

train.py

import torch.optim.optimizer
import torchvision
from model import *

from torch.utils.data import DataLoader

# 1.数据集准备
train_data = torchvision.datasets.CIFAR10('../dataset', train=True, transform=torchvision.transforms.ToTensor(),
                                          download=True)
test_data = torchvision.datasets.CIFAR10('../dataset', train=False, transform=torchvision.transforms.ToTensor(),
                                         download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
# print(train_data_size, test_data_size)  # 50000 10000

# 2.加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 3.创建网络模型
myModule = MyModule()

# 4.损失函数
loss_fn = nn.CrossEntropyLoss()
# 5.优化器
learningRate = 1e-2
optimizer = torch.optim.SGD(myModule.parameters(), lr=learningRate)
# 6.设置训练网络的一些参数
total_train_step = 0  # 记录训练的次数
total_test_step = 0  # 记录测试的次数
epoch = 10  # 记录训练的轮数
for i in range(epoch):
    print("-----第{}轮训练开始-----".format(i + 1))
    # 开始训练
    for data in train_dataloader:
        imgs, targets = data
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0:
            print("训练次数: {},loss: {}".format(total_train_step, loss.item()))
    # 测试步骤开始
    # 这一部分没有梯度,不需要再调优参数
    total_test_loss = 0
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            outputs = myModule(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss = total_test_loss + loss
    print("整体测试集上的Loss: {}".format(total_test_loss))

-----第1轮训练开始-----
训练次数: 100,loss: 2.298675060272217
训练次数: 200,loss: 2.285764694213867
训练次数: 300,loss: 2.2791736125946045
训练次数: 400,loss: 2.233513593673706
训练次数: 500,loss: 2.1184940338134766
训练次数: 600,loss: 2.019355297088623
训练次数: 700,loss: 2.0219309329986572
整体测试集上的Loss: 316.859619140625
-----第2轮训练开始-----
训练次数: 800,loss: 1.8966532945632935
训练次数: 900,loss: 1.8539228439331055
训练次数: 1000,loss: 1.9396780729293823
训练次数: 1100,loss: 1.9399535655975342
训练次数: 1200,loss: 1.6813435554504395
训练次数: 1300,loss: 1.6371924877166748
训练次数: 1400,loss: 1.744162678718567
训练次数: 1500,loss: 1.7939480543136597
整体测试集上的Loss: 296.5897216796875

1.6 使用tensorboard

目的：画出Loss曲线，Making sure gradient descent is working correctly. 

train.py

import torch.optim.optimizer
import torchvision
from torch.utils.tensorboard import SummaryWriter

from model import *

from torch.utils.data import DataLoader

# 1.数据集准备
train_data = torchvision.datasets.CIFAR10('../dataset', train=True, transform=torchvision.transforms.ToTensor(),
                                          download=True)
test_data = torchvision.datasets.CIFAR10('../dataset', train=False, transform=torchvision.transforms.ToTensor(),
                                         download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
# print(train_data_size, test_data_size)  # 50000 10000

# 2.加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 3.创建网络模型
myModule = MyModule()

# 4.损失函数
loss_fn = nn.CrossEntropyLoss()
# 5.优化器
learningRate = 1e-2
optimizer = torch.optim.SGD(myModule.parameters(), lr=learningRate)
# 6.设置训练网络的一些参数
total_train_step = 0  # 记录训练的次数
total_test_step = 0  # 记录测试的次数
epoch = 10  # 记录训练的轮数

# 添加tensorboard
writer = SummaryWriter('logs_train')

for i in range(epoch):
    print("-----第{}轮训练开始-----".format(i + 1))
    # 开始训练
    for data in train_dataloader:
        imgs, targets = data
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0:
            print("训练次数: {},loss: {}".format(total_train_step, loss.item()))
            # 画出损失函数
            writer.add_scalar('train_loss', loss.item(), total_train_step)
    # 测试步骤开始
    # 这一部分没有梯度,不需要再调优参数
    total_test_loss = 0
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            outputs = myModule(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss = total_test_loss + loss
    print("整体测试集上的Loss: {}".format(total_test_loss))
    total_test_step = total_test_step + 1
    writer.add_scalar('test_loss', total_test_loss, total_test_step)
writer.close()

1.7 保存每一轮的训练结果：

for i in range(epoch):
    print("-----第{}轮训练开始-----".format(i + 1))
    # 开始训练
    for data in train_dataloader:
        imgs, targets = data
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0:
            print("训练次数: {},loss: {}".format(total_train_step, loss.item()))
            # 画出损失函数
            writer.add_scalar('train_loss', loss.item(), total_train_step)
    # 测试步骤开始
    # 这一部分没有梯度,不需要再调优参数
    total_test_loss = 0
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            outputs = myModule(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss = total_test_loss + loss
    print("整体测试集上的Loss: {}".format(total_test_loss))
    total_test_step = total_test_step + 1
    writer.add_scalar('test_loss', total_test_loss, total_test_step)
    # 保存每一轮训练的结果
    torch.save(myModule, 'myModule_{}.pth'.format(i))
writer.close()

2. 计算整体的正确率

目的：查看训练的网络模型在测试集上的效果。

方法：预测正确的数量/整体的数量。

argmax：可以返回指定维度上最大值的索引。

test.py

import torch

outputs = torch.tensor([
    [0.1, 0.2],
    [0.3, 0.4]
])
# 填1的时候横向看,可以返回指定维度最大值的序号
preds = outputs.argmax(1)
targets = torch.tensor([0, 1])
print(preds == targets)
# 计算对应位置相等的个数
print((preds == targets).sum())

tensor([False,  True])
tensor(1)

for i in range(epoch):
    print("-----第{}轮训练开始-----".format(i + 1))
    # 开始训练
    for data in train_dataloader:
        imgs, targets = data
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0:
            print("训练次数: {},loss: {}".format(total_train_step, loss.item()))
            # 画出损失函数
            writer.add_scalar('train_loss', loss.item(), total_train_step)
    # 测试步骤开始
    # 这一部分没有梯度,不需要再调优参数
    total_test_loss = 0
    total_accuracy = 0  # 整体正确的个数
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            outputs = myModule(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss = total_test_loss + loss
            accuracy = (outputs.argmax(1) == targets).sum()
            total_accuracy = total_accuracy + accuracy
    print("整体测试集上的Loss: {}".format(total_test_loss))
    print("整体测试集上的正确率：{}".format(total_accuracy / test_data_size))
    writer.add_scalar('test_accuracy', total_accuracy / test_data_size, total_test_step)
    total_test_step = total_test_step + 1
    writer.add_scalar('test_loss', total_test_loss, total_test_step)
    # 保存每一轮训练的结果
    torch.save(myModule, './model_save/myModule_{}.pth'.format(i))
writer.close()

Files already downloaded and verified
Files already downloaded and verified
-----第1轮训练开始-----
训练次数: 100,loss: 2.291329860687256
训练次数: 200,loss: 2.2846291065216064
训练次数: 300,loss: 2.2555723190307617
训练次数: 400,loss: 2.145618438720703
训练次数: 500,loss: 2.0252487659454346
训练次数: 600,loss: 2.0127859115600586
训练次数: 700,loss: 1.9800595045089722
整体测试集上的Loss: 307.11065673828125
整体测试集上的正确率：0.2930999994277954
-----第2轮训练开始-----
训练次数: 800,loss: 1.8316386938095093
训练次数: 900,loss: 1.7961711883544922
训练次数: 1000,loss: 1.8935333490371704
训练次数: 1100,loss: 1.9779611825942993
训练次数: 1200,loss: 1.679609775543213
训练次数: 1300,loss: 1.6376134157180786
训练次数: 1400,loss: 1.707167148590088
训练次数: 1500,loss: 1.7555652856826782
整体测试集上的Loss: 292.63409423828125
整体测试集上的正确率：0.3312000036239624
-----第3轮训练开始-----
训练次数: 1600,loss: 1.7143256664276123
训练次数: 1700,loss: 1.6653105020523071
训练次数: 1800,loss: 1.942317247390747
训练次数: 1900,loss: 1.697310209274292
训练次数: 2000,loss: 1.8977160453796387
训练次数: 2100,loss: 1.532772183418274
训练次数: 2200,loss: 1.4647372961044312
训练次数: 2300,loss: 1.7696183919906616
整体测试集上的Loss: 257.3481750488281
整体测试集上的正确率：0.4074999988079071

......

3. 其他

3.1 train与eval

它们对一些网络模型、层是有作用的。

可以这样加入： 

3.2 使用GPU训练

图片来源于：b站up主 我是土堆

方法一：

train_gpu_1.py

import torch.optim.optimizer
import torchvision
from torch.utils.tensorboard import SummaryWriter
import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear

from torch.utils.data import DataLoader

# 1.数据集准备
train_data = torchvision.datasets.CIFAR10('../dataset', train=True, transform=torchvision.transforms.ToTensor(),
                                          download=True)
test_data = torchvision.datasets.CIFAR10('../dataset', train=False, transform=torchvision.transforms.ToTensor(),
                                         download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
# print(train_data_size, test_data_size)  # 50000 10000

# 2.加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 3.创建网络模型
# 搭建神经网络
class MyModule(nn.Module):
    def __init__(self):
        super(MyModule, self).__init__()
        self.model = nn.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(1024, 64),
            Linear(64, 10)
        )

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

myModule = MyModule()
if torch.cuda.is_available():
    myModule = myModule.cuda()

# 4.损失函数
loss_fn = nn.CrossEntropyLoss()
if torch.cuda.is_available():
    loss_fn = loss_fn.cuda()
# 5.优化器
learningRate = 1e-2
optimizer = torch.optim.SGD(myModule.parameters(), lr=learningRate)
# 6.设置训练网络的一些参数
total_train_step = 0  # 记录训练的次数
total_test_step = 0  # 记录测试的次数
epoch = 10  # 记录训练的轮数

# 添加tensorboard
writer = SummaryWriter('logs_train')

for i in range(epoch):
    print("-----第{}轮训练开始-----".format(i + 1))
    # 开始训练
    myModule.train()
    for data in train_dataloader:
        imgs, targets = data
        if torch.cuda.is_available():
            imgs = imgs.cuda()
            targets = targets.cuda()
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0:
            print("训练次数: {},loss: {}".format(total_train_step, loss.item()))
            # 画出损失函数
            writer.add_scalar('train_loss', loss.item(), total_train_step)
    # 测试步骤开始
    myModule.eval()
    # 这一部分没有梯度,不需要再调优参数
    total_test_loss = 0
    total_accuracy = 0  # 整体正确的个数
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            if torch.cuda.is_available():
                imgs = imgs.cuda()
                targets = targets.cuda()
            outputs = myModule(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss = total_test_loss + loss
            accuracy = (outputs.argmax(1) == targets).sum()
            total_accuracy = total_accuracy + accuracy
    print("整体测试集上的Loss: {}".format(total_test_loss))
    print("整体测试集上的正确率：{}".format(total_accuracy / test_data_size))
    writer.add_scalar('test_accuracy', total_accuracy / test_data_size, total_test_step)
    total_test_step = total_test_step + 1
    writer.add_scalar('test_loss', total_test_loss, total_test_step)
    # 保存每一轮训练的结果
    torch.save(myModule, './model_save/myModule_{}.pth'.format(i))
writer.close()

注：访问Google colaboratory， 可以免费使用GPU。

方法二：

import torch.optim.optimizer
import torchvision
from torch.utils.tensorboard import SummaryWriter
import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear

from torch.utils.data import DataLoader

# 定义训练的设备
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 1.数据集准备
train_data = torchvision.datasets.CIFAR10('../dataset', train=True, transform=torchvision.transforms.ToTensor(),
                                          download=True)
test_data = torchvision.datasets.CIFAR10('../dataset', train=False, transform=torchvision.transforms.ToTensor(),
                                         download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
# print(train_data_size, test_data_size)  # 50000 10000

# 2.加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 3.创建网络模型
# 搭建神经网络
class MyModule(nn.Module):
    def __init__(self):
        super(MyModule, self).__init__()
        self.model = nn.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(1024, 64),
            Linear(64, 10)
        )

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

myModule = MyModule()
myModule = myModule.to(device)

# 4.损失函数
loss_fn = nn.CrossEntropyLoss()
loss_fn = loss_fn.to(device)
# 5.优化器
learningRate = 1e-2
optimizer = torch.optim.SGD(myModule.parameters(), lr=learningRate)
# 6.设置训练网络的一些参数
total_train_step = 0  # 记录训练的次数
total_test_step = 0  # 记录测试的次数
epoch = 10  # 记录训练的轮数

# 添加tensorboard
writer = SummaryWriter('logs_train')

for i in range(epoch):
    print("-----第{}轮训练开始-----".format(i + 1))
    # 开始训练
    myModule.train()
    for data in train_dataloader:
        imgs, targets = data
        imgs = imgs.to(device)
        targets = targets.to(device)
        outputs = myModule(imgs)
        loss = loss_fn(outputs, targets)
        # 优化器优化模型
        # a.梯度清零
        optimizer.zero_grad()
        loss.backward()  # b.反向传播,拿到梯度
        optimizer.step()  # c.对参数进行优化
        total_train_step = total_train_step + 1
        if total_train_step % 100 == 0:
            print("训练次数: {},loss: {}".format(total_train_step, loss.item()))
            # 画出损失函数
            writer.add_scalar('train_loss', loss.item(), total_train_step)
    # 测试步骤开始
    myModule.eval()
    # 这一部分没有梯度,不需要再调优参数
    total_test_loss = 0
    total_accuracy = 0  # 整体正确的个数
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            imgs = imgs.to(device)
            targets = targets.to(device)
            outputs = myModule(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss = total_test_loss + loss
            accuracy = (outputs.argmax(1) == targets).sum()
            total_accuracy = total_accuracy + accuracy
    print("整体测试集上的Loss: {}".format(total_test_loss))
    print("整体测试集上的正确率：{}".format(total_accuracy / test_data_size))
    writer.add_scalar('test_accuracy', total_accuracy / test_data_size, total_test_step)
    total_test_step = total_test_step + 1
    writer.add_scalar('test_loss', total_test_loss, total_test_step)
    # 保存每一轮训练的结果
    torch.save(myModule, './model_save/myModule_{}.pth'.format(i))
writer.close()