KNN-分类算法的实现

    2010年的一道题，做了2天。。。对文件操作的处理，还是不够熟练。还得多熟悉才行，不然心里没底。

    从文本文件里读取训练样本的方法始终没有想出好的解决办法，目前很笨，先读取有多少行，就是说有多少个训练样本，再从头一行一行的读取到数组里。

    KNN的描述大致为：

    在一个文本文件里，每一行保存一个训练样本，有N_DIM个属性值，用空格分开，最后是一个分类，用“+”或“-”来表示。例如：

1.2 3.2 6.6 +

3.3 5.6 1.9 -

2.7 3.9 5.2 +

。。。

    待分类实例也有N_DIM个属性，分别与训练样本计算出距离值SIM，然后选出SIM值最小的前K个训练样本来，以这Ｋ个训练样本分类数多的作为待分类实例的分类值。

    SIM值是按向量的余弦值来计算，公式为sim=SUM(V1*Ｖ2) / [sqrt(V1*V1)*sqrt(V2*V2)]。

在 ＫＮＮ.h头文件里定义常量和相关读取数据、计算的函数：

#ifndef __HJDSAMPLE_INCLUDE__
#define __HJDSAMPLE_INCLUDE__

#define N_DIM 3 //每行3个属性值

#define N_KNN 3 //取最近的3个训练样本

//定义训练样本的结构

typedef struct SampleStruct
{
    double rProperties[N_DIM];
    char cType;
    double rSim; //距离
} SampleStructType;

//对训练样本数组进行快速排序，以便找到距离最小的前k个训练样本
void SampleStructType_qsort(SampleStructType ar_samples[], int nS, int nE)
{
    if(nS<nE)
    {
        double rKey = ar_samples[nS].rSim;
        //printf("\nqsort(%d, %d).\n", nS, nE);
        int i=nS, j=nE;
        while(i<j)
        {
            if(ar_samples[j].rSim<rKey)
            {
                if(ar_samples[i].rSim>rKey)
                {
                    SampleStructType_swap(&ar_samples[i], &ar_samples[j], N_DIM);
                    j--;
                    i++;
                }
                else
                {
                    i++;
                }
            }
            else
            {
                j--;
            }
        }
        int nMid =j;
        //printf("\nSuccess move round.\tnMid=%d\n", nMid);

        SampleStructType_swap(&ar_samples[nS], &ar_samples[j], N_DIM);
        //
        SampleStructType_qsort(ar_samples, nS, nMid);
        //
        SampleStructType_qsort(ar_samples, nMid+1, nE);
    }
}

//这里有点笨，先扫描一次训练样本的文本文件，得到有多少个训练样本
int getLineCount(const char * filename)
{
    int nLineCount=0;
    char ch;
    FILE *fp=fopen(filename, "r");
    if(NULL!=fp)
    {
        while((ch=fgetc(fp))!=EOF)
        {
            //printf("%c", ch);
            if(ch=='\n')
            {
                nLineCount++;
            }
        }
    }
    else
    {
        nLineCount=0;
    }
    fclose(fp);
    return(nLineCount);
}

//读取训练样本
int hjd_readSamples(const char * filename, SampleStructType *ar_samples, int nSamplesCount)
{
    int nScanRes = 0;
    FILE *fp=fopen(filename, "r");
    int nPos = 0;
    char ch='\0';
    if(NULL==fp)
    {
        nScanRes =-1;//cannot open file
    }
    else
    {
        for(nPos=0; nPos<nSamplesCount; nPos++)
        {
            for(int ndim=0; ndim<N_DIM; ndim++)
            {
                nScanRes = fscanf(fp, "%lf ", &ar_samples[nPos].rProperties[ndim]);
            }
            ch = fgetc(fp);//read catalog
            ar_samples[nPos].cType = ch;
        }
        nScanRes = nSamplesCount;
    }
    fclose(fp);
    return(nScanRes);
}

//读取待分类的实例
int hjd_readInstance(const char * filename, SampleStructType * objInstance)
{
    int nScanRes = 1;
    FILE *fp = fopen(filename, "r");
    //char ch='\0';
    if(NULL==fp)
    {
        nScanRes =-1;//cannot open file
    }
    else
    {
        for(int ndim=0; ndim<N_DIM; ndim++)
        {
            nScanRes = fscanf(fp, "%lf ", &(objInstance->rProperties)[ndim]);
        }
    }
    fclose(fp);
    return(nScanRes);
}

//计算两个向量的积
double vector_multiple(SampleStructType v1, SampleStructType v2, int nDIM)
{
    double rsum=0.0;
    for(int ndpos=0; ndpos<nDIM; ndpos++)
    {
        rsum = rsum + v1.rProperties[ndpos]*v2.rProperties[ndpos];
    }
    return(rsum);
}

double vector_abs(SampleStructType v1, int nDIM)
{
    double rsum=0.0;
    for(int ndpos=0; ndpos<nDIM; ndpos++)
    {
        rsum = rsum + v1.rProperties[ndpos]*v1.rProperties[ndpos];
    }
    rsum = sqrt(rsum);
    return(rsum);
}

//为了实现快速排序要用到的向量的交换
void SampleStructType_swap(SampleStructType  *s1, SampleStructType *s2, int nDIM)
{
    SampleStructType tmp;
    for(int nppos=0; nppos<nDIM; nppos++)
    {
        tmp.rProperties[nppos] = (s1->rProperties[nppos]);
    }
    tmp.cType = s1->cType;
    tmp.rSim = s1->rSim;

    for(int nppos=0; nppos<nDIM; nppos++)
    {
        s1->rProperties[nppos] = (s2->rProperties[nppos]);
    }
    s1->cType = s2->cType;
    s1->rSim = s2->rSim;

    for(int nppos=0; nppos<nDIM; nppos++)
    {
        s2->rProperties[nppos] = (tmp.rProperties[nppos]);
    }
    s2->cType = tmp.cType;
    s2->rSim = tmp.rSim;
}

//计算与训练样本的距离值
void knn_sim(SampleStructType ar_samples[], int nSampleCount, SampleStructType * objInstance)
{
    double rAbs2= vector_abs(*objInstance, N_DIM);
    //printf("\nSuccess calculate rAbs2=%lf.", rAbs2);
    for(int ncount=0; ncount<nSampleCount; ncount++)
    {
        double rSim = vector_multiple(ar_samples[ncount], (*objInstance), N_DIM);
        double rAbs1 = vector_abs(ar_samples[ncount], N_DIM);
        rSim = rSim/(rAbs1*rAbs2);
        ar_samples[ncount].rSim = rSim;
    }
    //after calculate every SIM value, sort the samples
    printf("\nSuccess calculate each rSim.\n");
    SampleStructType_qsort(ar_samples, 0, (nSampleCount-1));
    printf("\nSuccess sort the samples.\n");
    //count
    int nP=0, nM=0;
    for(int npos=0; npos<N_KNN; npos++)
    {
        if(ar_samples[npos].cType=='+')
            nP++;
        else
            nM++;
    }
    if(nP>nM)
        objInstance->cType='+';
    else
        objInstance->cType='-';
}
#endif

在主程序里调用方法就可以了，knn_app.c的内容为：

/*--------------------------------------------------------*/
/* Implement a K-Nearest-N Learning algrithom             */
/*--------------------------------------------------------*/
#include <stdio.h>
#include <stdlib.h>
#include <malloc.h>
#include <math.h>
#include <sys/time.h>
#include "hjdsample.h"

int main(void)
{
    int nSamplesCount = getLineCount("./samples.txt"); //记录有多少条训练样本
    SampleStructType ar_samples[nSamplesCount], objInstance;
    if((hjd_readSamples("./samples.txt", ar_samples, nSamplesCount)>0)&&(hjd_readInstance("./instances.txt", &objInstance)>0))
    {
        for(int nPos=0; nPos<nSamplesCount; nPos++)
        {
            for(int ndim=0; ndim<N_DIM; ndim++)
            {
                printf("%lf ", ar_samples[nPos].rProperties[ndim]);
            }
            printf("%c\n", ar_samples[nPos].cType);
        }
        printf("\nthe instance is:\n");
        for(int ndim=0; ndim<N_DIM; ndim++)
        {
            printf("%lf ", objInstance.rProperties[ndim]);
        }
        objInstance.cType='\0';
    }

    printf("\nStart to calculate.\n");
    knn_sim(ar_samples, nSamplesCount, &objInstance);
    printf("objInstance.cType=%c\n", objInstance.cType);
    return(0);
}

感觉真难呀，要在40分钟内写完这么多代码，我几乎做不到啊。。。