temp.cpp

// logistic-forest
// memo:書きなぐり
 
#include <iostream>
#include <cstdio>
#include <cmath>
#include <vector>
using namespace std;
 
double sigmoid(double x){return 1.0/(1.0+exp(-x));}

class ml_logistic{
    public: // todo:フィールドがら空き
    int dim;
    vector<double> weight;
    
    ml_logistic(int d):dim(d){
        weight.resize(d+1,0.0);
    }
    template<class Iterator>
    double test(Iterator id){
        double a;
        vector<double>::iterator iw=weight.begin();
        a=*iw;
        for (iw++;iw!=weight.end();iw++,id++)
            a+=*id**iw;
        return a;
    }
    template<class Iterator>
    void learn(Iterator id,bool cd,double eta){
        vector<double>::iterator iw=weight.begin();
        double b=(double)cd;
        double a=test(id);
        *iw-=eta*(sigmoid(a)-b);
        for (iw++;iw!=weight.end();iw++,id++)
            *iw-=eta*(sigmoid(a)-b)**id;
    }
    // for amp---
    template<class Iterator>
    double test(Iterator id,Iterator ia){
        double a;
        vector<double>::iterator iw=weight.begin();
        a=*ia**iw;
        for (iw++,ia++;iw!=weight.end();iw++,id++,ia++)
            a+=*ia**id**iw;
        return a;
    }
    template<class Iterator>
    void learn(Iterator id,bool cd,double eta,Iterator ia){
        vector<double>::iterator iw=weight.begin();
        double b=(double)cd;
        double a=test(id,ia);
        *iw-=eta*(sigmoid(a)-b);
        for (iw++;iw!=weight.end();iw++,id++)
            *iw-=eta*(sigmoid(a)-b)**id;
    }
};
 
int main(){
    int n,dim,asnum;
    vector<vector<double>> data;
    vector<bool> type;
    vector<vector<double>> weight;
    vector<double> amp;
    double eta=0.25;
    int i,j,t;
    vector<vector<double>>::iterator ivd;
    vector<bool>::iterator ib;
    vector<double>::iterator id;
    int cnt,lcnt,ac,bac;
 
    cin>>n>>dim>>asnum;
    data.resize(n);
    for (i=0;i<n;i++){
        cin>>t;type.push_back(t);
        for (j=0;j<dim;j++){
            cin>>t;data[i].push_back(t);
        }
    }
    weight.resize(asnum);
    for (i=0;i<asnum;i++)
        weight[i].resize(dim+1);
    
    vector<ml_logistic> mllg;
    for (i=0;i<asnum;i++)
        mllg.push_back(*(new ml_logistic(dim)));
    
    bac=0;
    
    for (cnt=0;cnt<100;cnt++){
        // init amp
        amp.assign(amp.size(),1.0);
        
        for (i=0;i<asnum;i++){
            for (lcnt=0;lcnt<10;lcnt++){
                for (vector<ml_logistic>::iterator ivml : mllg){
                    for (ivd=data.begin(),ib=type.begin();ivd!=data.end();ivd++,ib++){
                        mllg.learn((*ivd).begin(),*ib,eta,amp);
                    }
                }
            }
                
        }
            
        for (it=data.begin(),iu=type.begin();it!=data.end();it++,iu++){
            // learn
            mllg.learn((*it).begin(),*iu,eta);
        }
        ac=0;
        for (it=data.begin(),iu=type.begin();it!=data.end();it++,iu++){
            // test
            if ((mllg.test((*it).begin())>=0.0)==*iu) ac++;
        }
        
        if (ac>bac){
            weight.assign(mllg.weight.begin(),mllg.weight.end());
            bac=ac;
        }
        if (ac==n){cout<<"accept:"<<cnt<<endl;break;}
        if (1e-3<eta)eta*=0.9;
    }
    cout<<"bac:"<<bac<<endl;
    if (ac!=n){
        mllg.weight.assign(weight.begin(),weight.end());
    }
    
    for (iv=weight.begin();iv!=weight.end();iv++)
        printf("%8.4f ",*iv);
    cout<<endl;
    for (it=data.begin(),iu=type.begin();it!=data.end();it++,iu++){
        printf("%d %d\n",mllg.test((*it).begin())>0?1:0,(int)(*iu));
    }
 
    return 0;
}