plsa c++代码

#include<bits/stdc++.h>

using namespace std;

typedef long long ll;

const int INF=0x3f3f3f3f;
const int MAX_M=1e4+5;
const int MAX_K=30+5;
const int MAX_N=1e6+5;

struct node{
    int num;
    double top[MAX_K];
    node(){
        num=0;
        memset(top,0,sizeof(top));
    }
};

int M,K,N;//文档数、主题数、词数
string dic[MAX_N+1];//词表
map<string,int> mp;//词表map
double doc[MAX_M+1][MAX_K+1];//每篇文档的主题分布
double top[MAX_K+1][MAX_N+1];//每个主题的词分布
map<int,node> doc_words[MAX_M+1];//隐含变量，指示器

vector<string> data[MAX_M];//测试数据

int Rand(int n){//生成[0,n)之间的整合
    return (double)rand()/(RAND_MAX+1)*n;
}

void init(){
    for(int i=1;i<=M;i++){
        double s=0;
        for(int j=1;j<=K;j++){
            doc[i][j]=Rand(100)+1;
            s+=doc[i][j];
        }
        for(int j=1;j<=K;j++)doc[i][j]/=s;
    }
    for(int i=1;i<=K;i++){
        double s=0;
        for(int j=1;j<=N;j++){
            top[i][j]=Rand(100)+1;
            s+=top[i][j];
        }
        for(int j=1;j<=N;j++)top[i][j]/=s;
    }
}

void get_E(){
    for(int i=1;i<=M;i++){
        for(map<int,node>::iterator it=doc_words[i].begin();it!=doc_words[i].end();it++){
            int id=it->first;
            node &w=it->second;
            double s=0;
            for(int j=1;j<=K;j++)s+=doc[i][j]*top[j][id];
            for(int j=1;j<=K;j++)w.top[j]=doc[i][j]*top[j][id]/s;
        }
    }
}

void get_M(){
    for(int i=1;i<=K;i++){
        double s=0;
        fill(top[i],top[i]+N+1,0);
        for(int j=1;j<=M;j++){
            for(map<int,node>::iterator it=doc_words[j].begin();it!=doc_words[j].end();it++){
                int id=it->first;
                node &w=it->second;
                top[i][id]+=w.top[i]*w.num;
                s+=w.top[i]*w.num;
            }
        }
        for(int j=1;j<=N;j++){
            top[i][j]=top[i][j]/s;
        }
    }

    for(int i=1;i<=M;i++){
        double s=0;
        fill(doc[i],doc[i]+K+1,0);
        for(map<int,node>::iterator it=doc_words[i].begin();it!=doc_words[i].end();it++){
            node &w=it->second;
            for(int j=1;j<=K;j++){
                doc[i][j]+=w.top[j]*w.num;
                s+=w.top[j]*w.num;
            }
        }
        for(int j=1;j<=K;j++){
            doc[i][j]=doc[i][j]/s;
        }
    }
}

double get_MLE(){
    double res=0;
    for(int i=1;i<=M;i++){
        for(map<int,node>::iterator it=doc_words[i].begin();it!=doc_words[i].end();it++){
            int id=it->first;
            double p=0,c=it->second.num;
            for(int j=1;j<=K;j++)p+=doc[i][j]*top[j][id];
            res+=c*log(p);
        }
    }
    return res;
}

int kkk;
int rak[MAX_N+1];
bool cmp(int i,int j){
    return top[kkk][i]>top[kkk][j];
}

void show(){
    for(int i=1;i<=K;i++){
        cout<<i<<":";
        kkk=i;
        for(int j=1;j<=N;j++)rak[j]=j;
        sort(rak+1,rak+N+1,cmp);
        for(int j=1;j<=10;j++){
            int id=rak[j];
            cout<<dic[id]<<"*";
            printf("%.4f   ",top[i][id]);
        }
        cout<<endl;
    }
}

int main(){
    srand(time(NULL));
    freopen("C:\\Users\\28612\\Desktop\\data","r",stdin);
    M=0,K=20,N=0;
    cout<<"主题数K："<<K<<endl;
    int t;
    while(cin>>t){
        M++;
        while(t--){
            string s;
            cin>>s;
            data[M].push_back(s);
        }
    }
    cout<<"文档数M："<<M<<endl;

    for(int i=1;i<=M;i++){
        for(int j=0;j<data[i].size();j++){
            string s=data[i][j];
            if(mp[s]==0){
                dic[++N]=s;
                mp[s]=N;
            }
        }
    }
    cout<<"词汇数N："<<N<<endl;

    for(int i=1;i<=M;i++){
        for(int j=0;j<data[i].size();j++){
            string s=data[i][j];
            int id=mp[s];
            doc_words[i][id].num++;
        }
    }

    init();
    double last=-1e10;
    for(int i=1;i<=1000;i++){
        get_E();
        get_M();
        double p=get_MLE();
        cout<<i<<":"<<p<<endl;
        if(abs(p-last)<0.1)break;
        last=p;
    }
    show();
    return 0;
}

参考：
https://www.cnblogs.com/zhangchaoyang/articles/5668024.html