pytorch保存模型pth_Pytorch训练笔记

总结一些Pytorch训练模型的技巧。

指定使用哪些GPU

import os
os.environ["CUDA_VISIBLE_DEVICES"] = '0,1'

选择设备

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print('The device is: ', device)
model.to(device)

加载数据集，示例来自pytorch-tutorial

import torchvision
import torchvision.transforms as transforms

# Image preprocessing modules
transform = transforms.Compose([
    transforms.Pad(4),
    transforms.RandomHorizontalFlip(),
    transforms.RandomCrop(32),
    transforms.ToTensor()])

# CIFAR-10 dataset
train_dataset = torchvision.datasets.CIFAR10(root='../../data/',
                                             train=True, 
                                             transform=transform,
                                             download=True)

# Data loader
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                           batch_size=100, 
                                           shuffle=True)

如需自定义数据集，则需要继承torch.utils.data.dataset中的Dataset类

from torch.utils.data.dataset import Dataset
 
class MyCustomDataset(Dataset):
    def __init__(self, ...):
        # stuff
        
    def __getitem__(self, index):
        # stuff
        return (img, label)
 
    def __len__(self):
        return count

其中包括三个主要部分：

• __init__( ) 一些初始化过程写在这里
• __len__( ) 返回所有数据的数量
• __getitem__( ) 返回数据和标签，可以这样显式调用

img, label = MyCustomDataset.__getitem__(99)

更多内容可以查看自定义数据集的读取方法小结

保存，加载模型参数

# 保存整个网络
torch.save(model, PATH) 
# 保存网络中的参数, 速度快，占空间少
torch.save(model.state_dict(), PATH)
#--------------------------------------------------
# 对于上述保存方法，加载的方法分别是：
model_load = torch.load(PATH)

# 仅保存参数，加载时需要先得到网络结构
model_load = MyNet()
model_load.load_state_dict(torch.load(PATH))

如需保存更多信息，如优化器参数，模型准确率等

torch.save({'epoch': epoch + 1, 'state_dict': model.state_dict(), 
            'best_acc': train_accuracy,'optimizer': optimizer.state_dict()},PATH)

# 对应加载模型参数以字典的形式
model_dict = torch.load('./checkpoint.pth')
model.load_state_dict(model_dict['state_dict'])
optimizer.load_state_dict(model_dict['optimizer'])

微调模型，修改最后一层全连接层输出类别数，以适用于自己的数据集

fc_features = model.fc.in_features
# 假设自己数据集中有20类
model.fc = nn.Linear(fc_features, 20)

冻结模型参数

# 冻结1-8层模型参数，如使用block，则一个block为一层
ct = 1
for child in model.children():
    if ct <= 8:
        for param in child.parameters():
            param.requires_grad = False
    ct += 1
print('freeze model')

#对应的，优化器需要修改为
optimizer = torch.optim.SGD(filter(lambda p: p.requires_grad, model.parameters()), lr=learning_rate, momentum=0.9)

训练模型

# For updating learning rate
def update_lr(optimizer, lr):    
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr

# Train the model
total_step = len(train_loader)
curr_lr = learning_rate
for epoch in range(num_epochs):
    model.train()
    for i, (images, labels) in enumerate(train_loader):
        images = images.to(device)
        labels = labels.to(device)
        
        # Forward pass
        outputs = model(images)
        loss = criterion(outputs, labels)
        
        # Backward and optimize
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        
        if (i+1) % 100 == 0:
            print ("Epoch [{}/{}], Step [{}/{}] Loss: {:.4f}"
                   .format(epoch+1, num_epochs, i+1, total_step, loss.item()))

    # Decay learning rate
    if (epoch+1) % 20 == 0:
        curr_lr /= 3
        update_lr(optimizer, curr_lr)

测试模型，测试前需要调用model.eval()，如需在测试后继续训练，记得在训练前调用model.train()

# Test the model
model.eval()
with torch.no_grad():
    correct = 0
    total = 0
    for images, labels in test_loader:
        images = images.to(device)
        labels = labels.to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

    print('Accuracy of the model on the test images: {} %'.format(100 * correct / total))

使用多GPU训练，调用nn.DataParallel，来自知乎-pytorch如何使用多GPU

import torch.nn as nn

if device == torch.device('cuda'):
    print('set parallel')
    model_parallel = nn.DataParallel(model, device_ids=[0,3])
    torch.backends.cudnn.benchmark = True

# 对应的，保存模型时需要从model_parallel中取得model
torch.save(model_parallel.module.state_dict(), PATH)