pytorch中的优化器optimizer.param_groups用法

import torch
import torch.optim as optimh2
w1 = torch.randn(3, 3)
w1.requires_grad = True
w2 = torch.randn(3, 3)
w2.requires_grad = True
o = optim.Adam([w1])
print(o.param_groups)

[{'amsgrad': False,
  'betas': (0.9, 0.999),
  'eps': 1e-08,
  'lr': 0.001,
  'params': [tensor([[ 2.9064, -0.2141, -0.4037],
           [-0.5718,  1.0375, -0.6862],
           [-0.8372,  0.4380, -0.1572]])],
  'weight_decay': 0}]

Per the docs, the add_param_group method accepts a param_group parameter that is a dict. Example of use:h2import torch
import torch.optim as optimh2
w1 = torch.randn(3, 3)
w1.requires_grad = True
w2 = torch.randn(3, 3)
w2.requires_grad = True
o = optim.Adam([w1])
print(o.param_groups)
givesh2[{'amsgrad': False,
  'betas': (0.9, 0.999),
  'eps': 1e-08,
  'lr': 0.001,
  'params': [tensor([[ 2.9064, -0.2141, -0.4037],
           [-0.5718,  1.0375, -0.6862],
           [-0.8372,  0.4380, -0.1572]])],
  'weight_decay': 0}]
nowh2o.add_param_group({'params': w2})
print(o.param_groups)

[{'amsgrad': False,
  'betas': (0.9, 0.999),
  'eps': 1e-08,
  'lr': 0.001,
  'params': [tensor([[ 2.9064, -0.2141, -0.4037],
           [-0.5718,  1.0375, -0.6862],
           [-0.8372,  0.4380, -0.1572]])],
  'weight_decay': 0},
 {'amsgrad': False,
  'betas': (0.9, 0.999),
  'eps': 1e-08,
  'lr': 0.001,
  'params': [tensor([[-0.0560,  0.4585, -0.7589],
           [-0.1994,  0.4557,  0.5648],
           [-0.1280, -0.0333, -1.1886]])],
  'weight_decay': 0}]

# 动态修改学习率
for param_group in optimizer.param_groups:
    param_group["lr"] = lr 
# 得到学习率optimizer.param_groups[0]["lr"] h2# print('查看optimizer.param_groups结构：')
# i_list=[i for i in optimizer.param_groups[0].keys()]
# print(i_list)    
['amsgrad', 'params', 'lr', 'betas', 'weight_decay', 'eps']

# -*- coding: utf-8 -*-
#@Time    :2019/7/3 22:31
#@Author  :XiaoMa
 
from torch import nn as nn
import torch as t
from torch.autograd import Variable as V
#定义一个LeNet网络
class Net(nn.Module):
    def __init__(self):
        super(Net,self).__init__()
        self.features=nn.Sequential(
            nn.Conv2d(3,6,5),
            nn.ReLU(),
            nn.MaxPool2d(2,2),
            nn.Conv2d(6,16,5),
            nn.ReLU(),
            nn.MaxPool2d(2,3)
        )
        
        self.classifier=nn.Sequential(\
            nn.Linear(16*5*5,120),
            nn.ReLU(),
            nn.Linear(120,84),
            nn.ReLU(),
            nn.Linear(84,10)
            )
    def forward(self, x):
        x=self.features(x)
        x=x.view(-1,16*5*5)
        x=self.classifier(x)
        return x
net=Net()
 
from torch import optim #优化器
optimizer=optim.SGD(params=net.parameters(),lr=1)
optimizer.zero_grad()   #梯度清零，相当于net.zero_grad()
 
input=V(t.randn(1,3,32,32))
output=net(input)
output.backward(output)     #fake backward
optimizer.step()    #执行优化
 
#为不同子网络设置不同的学习率，在finetune中经常用到
#如果对某个参数不指定学习率，就使用默认学习率
optimizer=optim.SGD(
    [{'param':net.features.parameters()},    #学习率为1e-5
    {'param':net.classifier.parameters(),'lr':1e-2}],lr=1e-5
)
 
#只为两个全连接层设置较大的学习率，其余层的学习率较小
special_layers=nn.ModuleList([net.classifier[0],net.classifier[3]])
special_layers_params=list(map(id,special_layers.parameters()))
base_params=filter(lambda p:id(p) not in special_layers_params,net.parameters())
 
optimizer=t.optim.SGD([
    {'param':base_params},
    {'param':special_layers.parameters(),'lr':0.01}
],lr=0.001)

#调整学习率，新建一个optimizer
old_lr=0.1
optimizer=optim.SGD([
                {'param':net.features.parameters()},
                {'param':net.classifiers.parameters(),'lr':old_lr*0.5}],lr=1e-5)