LR模型属于广义线性模型,将特征空间映射成一种可能性。
损失函数:
逻辑回归函数:
算法流程:
代码实现:
设置迭代次数为500次,每次迭代,对每个训练文本都更新一次w
预测的方法是:先将训练文本与w相乘得到z,再判断g(z)是否大于等于0.5
(在PLA的代码基础上进行修改就可以得到LR)
数据集格式为:第一行为58个属性名称,最后一个是分享值(0或1),接下来n行是训练数据集,再继续来m行是预测数据集,其share值为空。需要求出预测数据集的share预测值。
<span style="font-size:14px;">#include<bits/stdc++.h>
#define rate 0.01
#define line_train 27751
#define line_test 11893
#define item_num 58
#define inf 1000000
using namespace std;
int line = line_train + line_test;
struct TEXT{
vector <double>feature;
int shares;
}text[40000];
void input()
{
string str;
double data;
ifstream fin("Datac_all.csv");
getline(fin, str);
for(int i=0;i<line;i++){
getline(fin, str);
istringstream ss(str);
vector<double> tmp;
while (!ss.eof()){
ss >> data; //忽略逗号
ss.ignore(str.size(), ',');
text[i].feature.push_back(data);
}
if(i<line_train){
text[i].shares = data;
}
}
fin.close();
cout << "Read done.\n";
}
void normalization(){
double MAX, MIN;
for(int i = 0; i < item_num; i++){
MAX = 0;
MIN = inf;
//求出Max-Min
for(int j=0;j<line;j++){
if(text[j].feature[i] > MAX)
MAX = text[j].feature[i];
if(text[j].feature[i] < MIN)
MIN = text[j].feature[i];
}
//归一化
for(int j = 0; j < line; j++){
if((MAX - MIN)!= 0)
text[j].feature[i] = (text[j].feature[i] - MIN) / (MAX - MIN);
else
text[j].feature[i] = 0;
}
}
}
double g(double z){
return 1.0/(1.0 + exp(-z));
}
double XW(vector<double> w, int i){
double A = w[0]*1.0;
for(int j = 1; j <= item_num; j++)
A += (double)w[j]*text[i].feature[j-1];
return A;
}
vector<double> new_w(vector<double> w, int index){//更新
vector<double> xe(item_num+1, 0);
double A = XW(w, index);
double err = text[index].shares - g(A);
w[0] += 1.0*rate*err;
for(int i = 1; i <= item_num; i++)
w[i] += rate * err * text[index].feature[i-1];
return w;
}
void print(vector<double> w){
for(int i = 0; i <= item_num; i++)
cout << w[i] << " ";
cout << endl << endl;
}
void LR(){
int DDL = 150;
vector<double> w(item_num+1, 1);
while(DDL--){ //迭代
for(int i = 0; i < line_train; i++){
w = new_w(w, i); //更新
}
}
print(w);
ofstream predict("LR.txt");
for(int i = line_train; i < line; i++){
double temp = XW(w, i); //预测
predict << (g(temp)>=0.5) << endl;
}
}
int main(){
input();
normalization();
LR();
cout << "Predict done" << endl;
return 0;
}</span>