《动手学深度学习(PyTorch版)》笔记3.2

注:书中对代码的讲解并不详细,本文对很多细节做了详细注释。另外,书上的源代码是在Jupyter Notebook上运行的,较为分散,本文将代码集中起来,并加以完善,全部用vscode在python 3.9.18下测试通过。

Chapter3 Linear Neural Networks

3.2 Implementations of Linear Regression from Scratch

复制代码
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import random
import torch
from d2l import torch as d2l

def synthetic_data(w, b, num_examples):  #@save
    """Generate y = Xw + b + noise."""
    #generates a random matrix X with dimensions (num_examples, len(w)) using a normal distribution with a mean of 0 and standard deviation of 1.
    X = torch.normal(0, 1, (num_examples, len(w)),dtype=torch.float32) 
    #calculates the target values y by multiplying the input matrix X with the weight vector w and adding the bias term b. 
    y = torch.matmul(X, w) + b  
    #And then adds some random noise to the target values y. The noise is generated from a normal distribution with mean 0 and standard deviation 0.01.                    
    y += torch.normal(0, 0.01, y.shape)             
    return X, y.reshape((-1, 1)) #The -1 in the first dimension means that PyTorch should automatically infer the size of that dimension based on the total number of elements. In other words, it is used to ensure that the reshaped tensor has the same total number of elements as the original tensor.

true_w=torch.tensor([2,-3.4],dtype=torch.float32)
true_b=4.2
features,labels=synthetic_data(true_w,true_b,1000)
print('features:',features[0],'\nlabel:',labels[0])

d2l.set_figsize()
d2l.plt.scatter(features[:,(1)].detach().numpy(),labels.detach().numpy(),1)
#plt.show()#显示散点图
#"features[:, 1]" selects the second column of the features tensor. 
#The detach() method is used to create a new tensor that shares no memory with the original tensor, and numpy() is then called to convert it to a NumPy array.
#"1" is the size of the markers in the scatter plot.

def data_iter(batch_size,features,labels):
    num_examples=len(features)
    indices=list(range(num_examples))
    #随机读取文本
    random.shuffle(indices)#"Shuffle the indices"意为打乱索引 
    for i in range(0,num_examples,batch_size):
        batch_indices=torch.tensor(indices[i:min(i+batch_size,num_examples)])#"min(i + batch_size, num_examples)" is used to handle the last batch, which might have fewer examples than batch_size.
        yield features[batch_indices],labels[batch_indices]

#初始化参数。从均值为0,标准差为0.01的正态分布中抽取随机数来初始化权重,并将偏置量置为0       
w=torch.normal(0,0.01,size=(2,1),requires_grad=True)
b=torch.zeros(1,requires_grad=True)

#定义线性回归模型
def linreg(X,w,b): #@save
    return torch.matmul(X,w)+b #广播机制:用一个向量加一个标量时,标量会加到向量的每一个分量上

#定义均方损失函数
def squared_loss(y_hat,y): #@save
    return (y_hat-y.reshape(y_hat.shape))**2/2

#定义优化算法:小批量随机梯度下降
def sgd(params,lr,batch_size): #@save
    with torch.no_grad():
        for param in params:
            param-=lr*param.grad/batch_size
            param.grad.zero_()

#轮数num_epochs和学习率lr都是超参数,先分别设为3和0.03,具体方法后续讲解
lr=0.03
num_epochs=3
batch_size=10

for epoch in range(num_epochs):
    for X,y in data_iter(batch_size,features,labels):
        l=squared_loss(linreg(X,w,b),y)
        l.sum().backward()#因为l是一个向量而不是标量,因此需要把l的所有元素加到一起来计算关于(w,b)的梯度
        sgd([w,b],lr,batch_size)
    with torch.no_grad():
        train_l=squared_loss(linreg(features,w,b),labels)
        print(f'epoch {epoch+1}:squared_loss {float(train_l.mean()):f}')
print(f'w的估计误差:{true_w-w.reshape(true_w.shape)}')
#结果中的grad_fn=<SubBackward0>表示这个tensor是由一个正向减法操作生成的
print(f'b的估计误差:{true_b-b}')#<RsubBackward1>表示由一个反向减法操作生成
相关推荐
我真的是大笨蛋2 小时前
K8S-Pod(下)
java·笔记·云原生·容器·kubernetes
昨日之日20062 小时前
Wan2.2-S2V - 音频驱动图像生成电影级质量的数字人视频 ComfyUI工作流 支持50系显卡 一键整合包下载
人工智能·音视频
纪元A梦2 小时前
贪心算法应用:化工反应器调度问题详解
算法·贪心算法
深圳市快瞳科技有限公司3 小时前
小场景大市场:猫狗识别算法在宠物智能设备中的应用
算法·计算机视觉·宠物
liulilittle3 小时前
OPENPPP2 —— IP标准校验和算法深度剖析:从原理到SSE2优化实现
网络·c++·网络协议·tcp/ip·算法·ip·通信
一个天蝎座 白勺 程序猿3 小时前
Python爬虫(47)Python异步爬虫与K8S弹性伸缩:构建百万级并发数据采集引擎
爬虫·python·kubernetes
XiaoMu_0015 小时前
基于Django+Vue3+YOLO的智能气象检测系统
python·yolo·django
SEO_juper5 小时前
大型语言模型SEO(LLM SEO)完全手册:驾驭搜索新范式
人工智能·语言模型·自然语言处理·chatgpt·llm·seo·数字营销
攻城狮7号5 小时前
腾讯混元翻译模型Hunyuan-MT-7B开源,先前拿了30个冠军
人工智能·hunyuan-mt-7b·腾讯混元翻译模型·30个冠军
superlls5 小时前
(算法 哈希表)【LeetCode 349】两个数组的交集 思路笔记自留
java·数据结构·算法