动手学深度学习v2 学习笔记01线性回归代码实现

手动实现代码

import random

import numpy as np
import torch
import matplotlib.pyplot as plt

def synthetic_data(w,b,num_examples):
    """ y = Xw+b+噪声 本次实际是y = x1*w1 + x2*w2 + b"""
    x = torch.normal(0,1,(num_examples,len(w)))##均值 标准差 num个样本 
    											##len第一维大小即w行数
    y = torch.matmul(x,w) + b ## 矩阵乘
    y +=torch.normal(0,0.01,y.shape) ## +b
    return x,y.reshape((-1,1)) ##-1表示自动推导

def data_iter(batch_size,features,labels):
    num_examples = len(features)
    indices = list(range(num_examples))
    random.shuffle(indices)##打乱
    for i in range(0,num_examples,batch_size):
        batch_indics = torch.tensor(indices[i:min(i+batch_size,num_examples)])
        yield features[batch_indics],labels[batch_indics] ##返回迭代器类似于指针，
        												##可以用for in 迭代访问

def linreg(X,w,b):
    return torch.matmul(X,w) + b##X,w矩阵乘 + b

def squared_loss(y_hat,y):
    """均方损失"""
    return (y_hat - y.reshape(y_hat.shape))**2 / 2## (y_hat - y)的平方/2 
    											##reshape为了防止广播
    ##个人理解广播就是两个矩阵不是相同形状时，自动扩展成相同形状再做加减

def sgd(params,lr,batch_size):
    """小批量随机梯度下降"""
    """我们想让它不断的往收敛方向走以找到较优解，而梯度的反方向就是最快的方向
    所以更新参数[w,b]"""
    with torch.no_grad():
        for param in params:
            param -= lr * param.grad / batch_size
            param.grad.zero_()

## 生成数据
true_w = torch.tensor([2,-3.4])
true_b = 4.2
features,labels = synthetic_data(true_w,true_b,1000)


lr = 0.03
num_epochs = 3
batch_size = 10
net = linreg
loss = squared_loss

w = torch.normal(0,0.01,size=(2,1),requires_grad=True)
b = torch.zeros(1,requires_grad=True)

for epoch in range(num_epochs):
    for X,y in data_iter(batch_size,features,labels):
        l = loss(net(X,w,b),y)
        l.sum().backward()
        sgd([w,b],lr,batch_size)
    with torch.no_grad():
        train_l = loss(net(features,w,b),labels)
        print(f'epoch{epoch+1},loss{float(train_l.mean()):f}')

print(f'w的估计误差：{true_w-w.reshape(true_w.shape)}')
print(f'b的估计误差：{true_b-b}')

##tensor转numpy  .detach().numpy()
plt.scatter(features[:,1].detach().numpy(),labels.detach().numpy(),1) ##画散点
x = np.linspace(-5,5,50)
y = x * w[1].detach().numpy() + b.detach().numpy()
plt.plot(x,y,color='green') ##画线
plt.show()
plt.scatter(features[:,0].detach().numpy(),labels.detach().numpy(),1)
y = x * w[0].detach().numpy() + b.detach().numpy()
plt.plot(x,y,color='green')
plt.show()

运行结果

pytorch实现代码

import numpy as np
import torch
from torch.utils import data
from torch import nn
import matplotlib.pyplot as plt

def synthetic_data(w,b,num_examples):
    """ y = Xw+b+噪声 本次实际是y = x1*w1 + x2*w2 + b"""
    x = torch.normal(0,1,(num_examples,len(w)))##均值 标准差 num个样本 
    											##len第一维大小即w行数
    y = torch.matmul(x,w) + b ## 矩阵乘
    y +=torch.normal(0,0.01,y.shape) ## +b
    return x,y.reshape((-1,1)) ##-1表示自动推导

def load_array(data_arrays,batch_size,is_train=True):
    """构造torch数据迭代器"""
    dataset = data.TensorDataset(*data_arrays)
    return data.DataLoader(dataset,batch_size,shuffle=is_train)

true_w = torch.tensor([2,-3.4])
true_b = 4.2
features,labels = synthetic_data(true_w,true_b,1000)

batch_size = 10
data_iter = load_array((features,labels),batch_size)##类似于手动实现时的yield返回迭代器

# print(next(iter(data_iter)))

net = nn.Sequential(nn.Linear(2,1))
net[0].weight.data.normal_(0,0.01)
net[0].bias.data.fill_(0)

loss = nn.MSELoss() ## 均方
trainer = torch.optim.SGD(net.parameters(),lr=0.03)##随机梯度下降，相同时间多次更新参数
                                                    ##加速收敛

num_epochs = 3
for epoch in range(num_epochs):
    for X,y in data_iter:
        trainer.zero_grad() ## 清零梯度，pytorch梯度累计而不是替换
                            ##个人觉得这句话也可以写在backward之后，
                            ##只要上次计算的梯度不影响下次就行
        l = loss(net(X),y)
        l.backward() ## 自动计算梯度
        trainer.step() ##根据梯度，使用优化器，更新参数
    l = loss(net(features),labels)## 这里只是看一下现在的loss
                                ## 不需要清梯度，没有backward，没有计算梯度
    print(f'epoch{epoch+1},loss{l:f}')##:f表示以浮点数形式输出，有点类似于c语言

##tensor转numpy  .detach().numpy()
plt.scatter(features[:,1].detach().numpy(),labels.detach().numpy(),1) ##画散点
x = np.linspace(-5,5,50)
y = x * net[0].weight.data[0][1].detach().numpy() + net[0].bias.data.detach().numpy()
plt.plot(x,y,color='green') ##画线
plt.show()
plt.scatter(features[:,0].detach().numpy(),labels.detach().numpy(),1)
y = x * net[0].weight.data[0][0].detach().numpy() + net[0].bias.data.detach().numpy()
plt.plot(x,y,color='green')
plt.show()

运行结果