神经网络（二）（pytorch）

优化算法1-梯度下降法

import numpy sa np
import torch
from torchvision.datasets import MNIST  #导入pytorch内置的mnist数据
from torch.utils.data import DataLoader
from torch import nn
from torch.autograd import Variable
import time
import matplotlib.pyplot as plt
%matplotlib inline

def data_tf(x):
    x = np.array(x,dtype='float32') / 255  #将数据变到0 ~ 1 之间
    x = (x - 0.5) / 0.5 #标准化
    x = x.reshape((-1,))
    x = torch.from_numpy(x)
    return x
trai_set = MNIST('./data',train=True,transform=data_tf,download=True)#载入数据集申明定义的数据变换
test_set = MNIST('./data',train=False,transform=data_tf,download=True)

#定义loss函数
criterion = nn.CrossEntropyLoss()

随机梯度下降法：

从0开始写：

def sgd_update(parameters,lr):
    for param in parameters:
        param.data = param.data - lr * param.grad.data

可以先把batch size设置为1

train_data = DataLoader(train_set,batch_size=1,shuffle=True)
#使用Sequential定义3层神经网络
net = nn.Sequential(
    nn.Linear(784,200),
    nn.ReLU(),
    nn.Linear(200,10),
)
#开始训练
losses1 = []
idx = 0
start = time.time()  #计时开始
for e in range(5):
    train_loss = 0
    for im,label in train_data:
        im = Variable(im)
        label = Variable(label)
        #前向传播
        net.zero_grad()
        loss.backward()
        sgd_update(net.parameters(),1e-2) #使用0.01的学习率
        #记录误差
        train_loss += loss.data[0]
        if idx % 30