• pytorch实现学习率衰减


    pytorch实现学习率衰减

    手动修改optimizer中的lr

    使用for循环来手动人工进行learning rate的衰减

    model = net()
    LR = 0.01
    optimizer = Adam(model.parameters(),lr = LR)
    lr_list = []
    for epoch in range(100):
        if epoch % 5 == 0:
            for p in optimizer.param_groups:
                p['lr'] *= 0.9#注意这里
        lr_list.append(optimizer.state_dict()['param_groups'][0]['lr'])
    plt.plot(range(100),lr_list,color = 'r')
    
    image-20200914211815982

    使用lr_scheduler

    LambdaLR——lambda函数衰减

    torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda, last_epoch=-1)
    

    lr_lambda 会接收到一个int参数:epoch,然后根据epoch计算出对应的lr。如果设置多个lambda函数的话,会分别作用于Optimizer中的不同的params_group

    import numpy as np 
    lr_list = []
    model = net()
    LR = 0.01
    optimizer = Adam(model.parameters(),lr = LR)
    
    lambda1 = lambda epoch:np.sin(epoch) / epoch
    scheduler = lr_scheduler.LambdaLR(optimizer,lr_lambda = lambda1)
    
    for epoch in range(100):
        scheduler.step()
        lr_list.append(optimizer.state_dict()['param_groups'][0]['lr'])
    plt.plot(range(100),lr_list,color = 'r')
    
    image-20200914212214994

    StepLR——阶梯式衰减

    torch.optim.lr_scheduler.StepLR(optimizer, step_size, gamma=0.1, last_epoch=-1)
    

    每个一定的epoch,lr会自动乘以gamma进行阶梯式衰减

    ⭐⭐⭐注意:pytorch1.1.0之后scheduler.step()要放在optimizer.step()之后!!!

    lr_list = []
    model = net()
    LR = 0.01
    optimizer = Adam(model.parameters(),lr = LR)
    scheduler = lr_scheduler.StepLR(optimizer,step_size=5,gamma = 0.8)
    for epoch in range(100):
        scheduler.step()
        lr_list.append(optimizer.state_dict()['param_groups'][0]['lr'])
    plt.plot(range(100),lr_list,color = 'r')
    
    image-20200914213252927

    MultiStepLR——多阶梯式衰减

    torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones, gamma=0.1, last_epoch=-1)
    

    三段式lr,epoch进入milestones范围内即乘以gamma,离开milestones范围之后再乘以gamma。这种衰减方式也是在学术论文中最常见的方式,一般手动调整也会采用这种方法。

    lr_list = []
    model = net()
    LR = 0.01
    optimizer = Adam(model.parameters(),lr = LR)
    scheduler = lr_scheduler.MultiStepLR(optimizer,milestones=[20,80],gamma = 0.9)
    for epoch in range(100):
        scheduler.step()
        lr_list.append(optimizer.state_dict()['param_groups'][0]['lr'])
    plt.plot(range(100),lr_list,color = 'r')
    
    image-20200914213738384

    ExponentialLR——指数连续衰减

    torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma, last_epoch=-1)
    

    每个epoch中lr都乘以gamma

    lr_list = []
    model = net()
    LR = 0.01
    optimizer = Adam(model.parameters(),lr = LR)
    scheduler = lr_scheduler.ExponentialLR(optimizer, gamma=0.9)
    for epoch in range(100):
        scheduler.step()
        lr_list.append(optimizer.state_dict()['param_groups'][0]['lr'])
    plt.plot(range(100),lr_list,color = 'r')
    
    image-20200914214232247

    CosineAnnealingLR——余弦退火衰减

     torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max, eta_min=0, last_epoch=-1)
    

    T_max 对应1/2个cos周期所对应的epoch数值

    eta_min 为最小的lr值,默认为0

    image-20200914214534397

    ReduceLROnPlateau

    torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=0.1, patience=10, verbose=False, threshold=0.0001, threshold_mode='rel', cooldown=0, min_lr=0, eps=1e-08)
    

    在发现loss不再降低或者acc不再提高之后,降低学习率。

    各参数意义如下:

    mode:'min'模式检测metric是否不再减小,'max'模式检测metric是否不再增大;

    factor: 触发条件后lr*=factor;

    patience:不再减小(或增大)的累计次数;

    verbose:触发条件后print;

    threshold:只关注超过阈值的显著变化;

    threshold_mode:有rel和abs两种阈值计算模式,rel规则:max模式下如果超过best(1+threshold)为显著,min模式下如果低于best(1-threshold)为显著;abs规则:max模式下如果超过best+threshold为显著,min模式下如果低于best-threshold为显著;

    cooldown:触发一次条件后,等待一定epoch再进行检测,避免lr下降过速;

    min_lr:最小的允许lr;

    eps:如果新旧lr之间的差异小与1e-8,则忽略此次更新。

    参考:https://www.jianshu.com/p/9643cba47655

  • 相关阅读:
    hdu 4638 Group 线段树
    hdu 4635 Strongly connected 强连通分量
    hdu 4604 Deque
    hdu 1000 A + B Problem
    数组相关
    文本文件相关
    硬件电路中VCC,VDD,VEE,VSS有什么区别
    VIPM链接LabVIEW问题
    Touch实现轻扫
    touchesMoved 实现拖拽
  • 原文地址:https://www.cnblogs.com/Jason66661010/p/13669683.html
Copyright © 2020-2023  润新知