class torch.optim.lr_scheduler.CosineAnnealingLR

参考链接: class torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max, eta_min=0, last_epoch=-1, verbose=False)
配套代码下载链接: 测试学习率调度器.zip

实验代码展示:

# torch.optim.lr_scheduler.CosineAnnealingLR

import matplotlib.pyplot as plt
import numpy as np 
import torch
from torch.utils.data import Dataset, DataLoader
from torch import nn
from torch.autograd import Function
import random
import os
seed = 20200910
os.environ['PYTHONHASHSEED'] = str(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)  # if you are using multi-GPU.
np.random.seed(seed)  # Numpy module.
random.seed(seed)  # Python random module.
torch.manual_seed(seed)
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True

class Dataset4cxq(Dataset):
    def __init__(self, length):
        self.length = length

    def __len__(self):
        return self.length
        
    def __getitem__(self, index):
        if type(index) != type(2) and type(index) != (slice):
           raise  TypeError('索引类型错误,程序退出...')
        
        # index 是单个数
        if type(index) == type(2):
            if index >= self.length or index < -1 * self.length:
                # print("索引越界,程序退出...")
                raise IndexError("索引越界,程序退出...")
            elif index < 0:
                index = index + self.length 
            
            Celsius = torch.randn(1,1,dtype=torch.float).item()
            Fahrenheit = 32.0 + 1.8 * Celsius
            return Celsius, Fahrenheit 
    
        
def collate_fn4cxq(batch):
    list_c = []
    list_f = []
    for c, f in batch:
        list_c.append(c)
        list_f.append(f)
    list_c = torch.tensor(list_c)
    list_f = torch.tensor(list_f)
    return list_c, list_f



if __name__ == "__main__":
    my_dataset = Dataset4cxq(32)
    # for c,f in my_dataset:
    #     print(type(c),type(f))
    dataloader4cxq = torch.utils.data.DataLoader(
        dataset=my_dataset, 
        batch_size=8,
        # batch_size=2,
        drop_last=True,
        # drop_last=False,
        shuffle=True,  #  True   False
        # shuffle=False,  #  True   False
        collate_fn=collate_fn4cxq,
        # collate_fn=None,
    )

    # for cnt, data in enumerate(dataloader4cxq, 0):
    #     # pass
    #     sample4cxq, label4cxq = data
    #     print('sample4cxq的类型: ',type(sample4cxq),'\tlabel4cxq的类型: ',type(label4cxq))
    #     print('迭代次数:', cnt, '  sample4cxq:', sample4cxq, '  label4cxq:', label4cxq)

    
    
    
    
    print('开始创建模型'.center(80,'-'))
    model = torch.nn.Linear(in_features=1, out_features=1, bias=True)  # True # False
    model.cuda()
    # optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
    optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
    # 模拟华氏度与摄氏度之间的转换  
    # Fahrenheit = 32 + 1.8 * Celsius
    model.train()
    cost_function = torch.nn.MSELoss()
    epochs = 100001  # 100001
    epochs = 10001  # 100001
    print('\n')
    print('开始训练模型'.center(80,'-'))
    list4delta = list()
    list4epoch = list()
    
    # scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=(lambda epoch: 0.99 ** (epoch//1000)))
    scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=1000)
    
    for epoch in range(epochs):
        # with torch.no_grad():
        #     Celsius = torch.randn(10,1,dtype=torch.float).cuda()
        #     Fahrenheit = 32.0 + 1.8 * Celsius
        #     Fahrenheit = Fahrenheit.cuda()

        # Celsius = torch.randn(1,1,dtype=torch.float,requires_grad=False).cuda()  # requires_grad=False  True
        # Fahrenheit = 32.0 + 1.8 * Celsius
        # Fahrenheit = Fahrenheit.cuda()        # requires_grad=False
        total_loss = 0.0
        for cnt, data in enumerate(dataloader4cxq, 0):
            Celsius, Fahrenheit = data
            Celsius, Fahrenheit = Celsius.cuda().view(-1,1), Fahrenheit.cuda().view(-1,1)
            output = model(Celsius)
            loss = cost_function(output, Fahrenheit)
            total_loss += loss.item()
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
        scheduler.step()
            
        if epoch % 100 == 0:  # if epoch % 1000 == 0:
            list4delta.append(total_loss)
            list4epoch.append(epoch)
            
        if epoch % 500 == 0:
            info = '\nepoch:{0:>6}/{1:<6}\t'.format(epoch,epochs)
            for k, v in model.state_dict().items():
                info += str(k)+ ':' + '{0:<.18f}'.format(v.item()) + '\t'
                # info += str(k)+ ':' + str(v.item()) + '\t'
            print(info)

    fig, ax = plt.subplots() 
    # ax.plot(10*np.random.randn(100),10*np.random.randn(100),'o')
    ax.plot(list4epoch, list4delta, 'r.-', markersize=8)
    ax.set_title("Visualization For My Model's Errors")
    plt.show()

控制台下结果输出:

Windows PowerShell
版权所有 (C) Microsoft Corporation。保留所有权利。

尝试新的跨平台 PowerShell https://aka.ms/pscore6

加载个人及系统配置文件用了 1009 毫秒。
(base) PS C:\Users\chenxuqi\Desktop\News4cxq\测试学习率调度器>  & 'D:\Anaconda3\envs\pytorch_1.7.1_cu102\python.exe' 'c:\Users\chenxuqi\.vscode\extensions\ms-python.python-2021.1.502429796\pythonFiles\lib\python\debugpy\launcher' '50296' '--' 'c:\Users\chenxuqi\Desktop\News4cxq\测试学习率调度器\test11.py'
-------------------------------------开始创建模型-------------------------------------


-------------------------------------开始训练模型-------------------------------------

epoch:     0/10001      weight:0.962383031845092773     bias:0.980020046234130859       

epoch:   500/10001      weight:1.088732838630676270     bias:2.599665641784667969       

epoch:  1000/10001      weight:1.105252861976623535     bias:2.957001924514770508       

epoch:  1500/10001      weight:1.126199364662170410     bias:3.319435834884643555       

epoch:  2000/10001      weight:1.240598320960998535     bias:4.924868583679199219

epoch:  2500/10001      weight:1.289438247680664062     bias:6.509951591491699219

epoch:  3000/10001      weight:1.319561481475830078     bias:6.864211082458496094

epoch:  3500/10001      weight:1.351148366928100586     bias:7.226140975952148438

epoch:  4000/10001      weight:1.468982815742492676     bias:8.826252937316894531

epoch:  4500/10001      weight:1.528319001197814941     bias:10.403023719787597656

epoch:  5000/10001      weight:1.521693468093872070     bias:10.755671501159667969

epoch:  5500/10001      weight:1.527165293693542480     bias:11.117164611816406250

epoch:  6000/10001      weight:1.597053766250610352     bias:12.710453033447265625

epoch:  6500/10001      weight:1.665725708007812500     bias:14.275875091552734375

epoch:  7000/10001      weight:1.663899421691894531     bias:14.626576423645019531

epoch:  7500/10001      weight:1.688347816467285156     bias:14.987486839294433594

epoch:  8000/10001      weight:1.756373763084411621     bias:16.570549011230468750

epoch:  8500/10001      weight:1.752237319946289062     bias:18.118822097778320312

epoch:  9000/10001      weight:1.768874168395996094     bias:18.466447830200195312

epoch:  9500/10001      weight:1.757285594940185547     bias:18.826288223266601562

epoch: 10000/10001      weight:1.771663427352905273     bias:20.393590927124023438

运行结果截图展示: