(十)pytorch多线程训练，DataLoader的num_works参数设置

一、概述

数据集较小时（小于2W）建议num_works不用管默认就行，因为用了反而比没用慢。
当数据集较大时建议采用，num_works一般设置为（CPU线程数+-1）为最佳，可以用以下代码找出最佳num_works（注意windows用户如果要使用多核多线程必须把训练放在if __name__ == '__main__':下才不会报错）

二、代码

import time
import torch.utils.data as d
import torchvision
import torchvision.transforms as transforms
 
 
if __name__ == '__main__':
    BATCH_SIZE = 100
    transform = transforms.Compose([transforms.ToTensor(),
                                    transforms.Normalize((0.5,), (0.5,))])
    train_set = torchvision.datasets.MNIST('\mnist', download=False, train=True, transform=transform)
    
    # data loaders
    train_loader = d.DataLoader(train_set, batch_size=BATCH_SIZE, shuffle=True)
    
    for num_workers in range(20):
        train_loader = d.DataLoader(train_set, batch_size=BATCH_SIZE, shuffle=True, num_workers=num_workers)
        # training ...
        start = time.time()
        for epoch in range(1):
            for step, (batch_x, batch_y) in enumerate(train_loader):
                pass
        end = time.time()
        print('num_workers is {} and it took {} seconds'.format(num_workers, end - start))

 三、查看线程数

1、cpu个数

grep 'physical id' /proc/cpuinfo | sort -u

2、核心数

grep 'core id' /proc/cpuinfo | sort -u | wc -l

3、线程数

grep 'processor' /proc/cpuinfo | sort -u | wc -l

4、例子

命令执行结果如图所示，根据结果得知，此服务器有1个cpu，6个核心，每个核心2线程，共12线程。