SVM.rar_SVM 测试集_SVM分类_SVM数据分类_svm分类器_测试数据

// SVM.cpp : 定义控制台应用程序的入口点。 // #include "stdafx.h" #include <cassert> #include <stdlib.h> #include <stdio.h> #include <time.h> #include <memory.h> void GetDataFromFile(DATAINFO* pData, const char *pFileName, const int nDataLength); void GetSVMVector(DATAINFO* pData, int nDataLength, double* pVector, int nVectorLength); void ClassTestData(DATAINFO* pData, int nDataLength, double* pVector, int nVectorLength, int* pTestType, DATASCORE *pDataScore); void StatistClassResult(DATAINFO* pData, int nDataLength, int* pTestType, float *pResult); float max(float a, float b); float min(float a, float b); void sort(DATASCORE *pDataScore, int nDataLength); float GetAUC(DATASCORE *pDataScore, int nDataLength); int _tmain(int argc, _TCHAR* argv[]) { double dwVector[NUM_FEATURE] = {0}; int nTestTypeBuf[NUM_TEST_DATA] = {0};//测试样本数 DATAINFO *pData = new DATAINFO[NUM_TRAIN_DATA]; assert(pData); const char strTrainFileName[] = "train.data"; //const char strTrainFileName[] = "1.data"; const char strTestFileName[] = "test.data"; GetDataFromFile(pData, strTrainFileName, NUM_TRAIN_DATA); GetSVMVector(pData, NUM_TRAIN_DATA, dwVector, NUM_FEATURE); if(pData) delete[] pData; pData = new DATAINFO[NUM_TEST_DATA]; DATASCORE *pDataScore = new DATASCORE[NUM_TEST_DATA]; assert(pData && pDataScore); GetDataFromFile(pData, strTestFileName, NUM_TEST_DATA); ClassTestData(pData, NUM_TEST_DATA, dwVector, NUM_FEATURE, nTestTypeBuf, pDataScore); sort(pDataScore, NUM_TEST_DATA); float fResult[3] = {0}; StatistClassResult(pData, NUM_TEST_DATA, nTestTypeBuf, fResult); printf("SVM分类器系数如下：\n"); for(int i = 0; i < NUM_FEATURE; i++) { printf("W[%d] = %f\n", i, dwVector[i]); } printf("True Positive rate：%f\nFalse Positive rate：%f\n总的分类正确率:%f\n", fResult[0], fResult[1], fResult[2]); // printf("Min score = %f\nMax score = %f\n", pDataScore[0].score, pDataScore[NUM_TEST_DATA - 1].score); printf("分类器AUC得分：%f\n", GetAUC(pDataScore, NUM_TEST_DATA)); if(pData) { delete[] pData; pData = NULL; } if (pDataScore) { delete[] pDataScore; pDataScore = NULL; } printf("Please input enter to quit!\n"); getchar(); return 0; } //从文件中读取样本数据（训练集和测试集都用此函数） //pData：存储读取到的数据的指针 //pFileName：存储数据的文件名 //nDataLength：样本数目 void GetDataFromFile(DATAINFO* pData, const char *pFileName, const int nDataLength) { assert(pData&&pFileName); FILE *pFile = fopen(pFileName, "r"); assert(pFile); int i = 0; char strLine[128] = {0}; char strTemp[16] = {0}; while(i < nDataLength) { memset(strLine, 0, 128); fgets(strLine, 128, pFile); int j = 0, k = 0, m = 0; while(0 != strLine[j]) { if(' ' == strLine[j] || 0X0A == strLine[j]) { strTemp[k] = 0; if(NUM_FEATURE == m) { pData[i].nType = atoi(strTemp); m = 0; } else { pData[i].dwFeature[m] = atof(strTemp); m++; } //ZeroMemory(strTemp, 16); memset(strTemp, 0, 16); j++; k = 0; } else { strTemp[k++] = strLine[j++]; } } i++; } if(pFile) { fclose(pFile); pFile = NULL; } } //根据训练集样本得到SVM分类器系数函数 //pData：训练样本的数据指针 //nDataLength：训练样本数目 //pVector：SVM系数存储指针 //nVectorLength：SVM训练器系数数目（10） void GetSVMVector(DATAINFO* pData, int nDataLength, double* pVector, int nVectorLength) { assert(pData && pVector); float *pfWeight = new float[nDataLength]; assert(pfWeight); float fLow, fHigh, fSum = 0.0, fOld; const float fC = 0.5; float fError1 = 0, fError2 = 0; float fg1 = 0, fg2 = 0; int nNum1, nNum2; memset(pfWeight, 0, sizeof(float) * nDataLength); srand((unsigned)time(NULL)); for(int i = 0; i < NUM_ITERATAR; i++) { fg1 = 0, fg2 = 0, fSum = 0; nNum1 = rand() % nDataLength; nNum2 = rand() % nDataLength; if(nNum1 == nNum2) continue; if(pData[nNum1].nType == pData[nNum2].nType) { fLow = max(0, pfWeight[nNum2] + pfWeight[nNum1] - fC); fHigh = min(fC, pfWeight[nNum2] + pfWeight[nNum1]); } else { fLow = max(0, pfWeight[nNum2] - pfWeight[nNum1]); fHigh = min(fC, fC + pfWeight[nNum2] - pfWeight[nNum1]); } for(int j = 0; j < NUM_FEATURE; j++) { fSum += 2.0 * pData[nNum1].dwFeature[j] * pData[nNum2].dwFeature[j] - pData[nNum1].dwFeature[j] * pData[nNum1].dwFeature[j] - pData[nNum2].dwFeature[j] * pData[nNum2].dwFeature[j]; //fSum += (pData[nNum1].dwFeature[j] - pData[nNum2].dwFeature[j]) * (pData[nNum1].dwFeature[j] - pData[nNum2].dwFeature[j]); fg1 = (float)(pVector[j] * pData[nNum1].dwFeature[j]); fg2 = (float)(pVector[j] * pData[nNum2].dwFeature[j]); } fg1 += NUM_B_THRESHOLD; fg2 += NUM_B_THRESHOLD; fOld = pfWeight[nNum2]; pfWeight[nNum2] -= pData[nNum2].nType * (fError1 - fError2) / fSum; float ff = pfWeight[nNum2]; fError1 = fg1 - pData[nNum1].nType; fError2 = fg2 - pData[nNum2].nType; if(pfWeight[nNum2] >= fHigh) { pfWeight[nNum2] = fHigh; } else if(pfWeight[nNum2] <= fLow) { pfWeight[nNum2] = fLow; } pfWeight[nNum1] += pData[nNum1].nType * pData[nNum2].nType * (fOld - pfWeight[nNum2]); for(int j = 0; j < NUM_FEATURE; j++) { pVector[j] = 0; for(int k = 0; k < nDataLength; k++) { pVector[j] += pfWeight[k] * pData[k].nType * pData[k].dwFeature[j]; } } if(0 == (i + 1)%1000) { printf("已经迭代%d次！\n", i + 1); } } if(pfWeight) { delete[] pfWeight; pfWeight = NULL; } } //对测试集进行分类 //pData：测试集样本数据指针 //nDataLength：测试集样本总数 //pVector：SVM分类器系数指针 //nVectorLength：SVM训练器系数数目（10） //pTestType：存放测试集分类类型的指针 //pDataScore：存储测试集样本得分及测试集样本类型的指针 void ClassTestData(DATAINFO* pData, int nDataLength, double* pVector, int nVectorLength, int* pTestType, DATASCORE *pDataScore) { assert(pData && pVector && pTestType); double dwSum = 0; for(int i = 0; i < nDataLength; i++) { for(int j = 0; j < nVectorLength; j++) { dwSum += pVector[j] * pData[i].dwFeature[j] ; } dwSum += NUM_B_THRESHOLD; if( dwSum > 0) { pTestType[i] = 1; } else { pTestType[i] = -1; } pDataScore[i].score = dwSum; pDataScore[i].nType = pData[i].nType; } } //统计分类器分类正确率 //pData：测试集样本数据指针 //nDataLength：测试集样本总数 //pVector：SVM分类器系数指针 //pResult：存放分类正确率 pResult[0]（True Positive rate） pResult[1]（False Positive rate） //pResult[2]（总的分类正确率） void StatistClassResult(DATAINFO* pData, int nDataLength, int* pTestType, float *pResult) { assert(pData && pTestType); int sum1 = 0, sum2 = 0, nType1Num = 0; for(int i = 0; i < nDataLength; i++) { if(pTestType[i] == pData[i].nType) { if(1 == pTestType[i]) { sum1++; } else { sum2++; } } if(1 == pData[i].nType) nType1Num++; } pResult[0] = (float)sum1/nType1Num; pResult[1] = (float)sum2/(nDataLength - nType1Num); pResult[2] = (float)(sum1 + sum2)/nDataLength; } //求二者中较大者 float max(float a, float b) { return a > b ? a : b; } //求二者中较小者 float min(float a, float b) { return a < b ? a : b; } //对测试集样本得分进行排序，用于AUC的计算 //pDataScore：测试集样本得分存储指针 //nDataLength：测试集样本总数 void sort(DATASCORE *pDataScore, int nDataLength) { assert(pDataScore); DATASCORE ScoreTemp; for(int i = 0; i < nDataLength; i++) { for(int j = i + 1; j < nDataLength; j++) { if(pDataScore[i].score > pDataScore[j].score) { ScoreTemp = pDataScore[i]; pDataScore[i] = pDataScore[j]; pDataScore[j] = ScoreTemp; } } } } //计算AUC得分 //pDataScore：测试集样本得分存储指针 //nDataLength：测试集样本总数 float GetAUC(DATASCOR