212 lines
6.2 KiB
C++
212 lines
6.2 KiB
C++
/*********************************************************************************************
|
||
Description: 用高斯混合模型进行曲线的拟合
|
||
|
||
Copyright : All right reserved by ZheYuan.BJ
|
||
|
||
Author : Zhang Zhonghai
|
||
Date : 2023/09/12
|
||
***********************************************************************************************/
|
||
|
||
#include <iostream>
|
||
#include <iomanip>
|
||
#include <fstream>
|
||
#include <sstream>
|
||
#include <algorithm>
|
||
#include <random>
|
||
#include <unordered_map>
|
||
#include <omp.h>
|
||
#include <time.h>
|
||
#include <string>
|
||
#include <vector>
|
||
#include <queue>
|
||
#include <filesystem>
|
||
#include <thread>
|
||
#ifdef _WIN32
|
||
#include <io.h>
|
||
#include <process.h>
|
||
#define F_OK 0
|
||
#else
|
||
#include <unistd.h>
|
||
#endif
|
||
#include <mat.h>
|
||
#include "gmm.h"
|
||
#include "thread_pool.h"
|
||
#include "CommonLib/matlab_io.h"
|
||
using namespace std;
|
||
using std::cout;
|
||
using std::vector;
|
||
namespace fs = std::filesystem;
|
||
|
||
/* 将x向量放到宽度为binWidth大小的桶中，功能类似matlab的hist*/
|
||
void PutXtoBin(double* x, int xSize, double binWidth, vector<double>& vXBin, vector<double>& vYBin) {
|
||
double maxX = 0.0;
|
||
for (int i = 0; i < xSize; ++i) {
|
||
if (maxX < x[i]) maxX = x[i];
|
||
}
|
||
int binSize = (int)((maxX + binWidth / 2) / binWidth + 1);
|
||
double binMaxVal = (binSize - 1) * binWidth;
|
||
if (binMaxVal > maxX) { // 确保最后一个bin不大于maxX，而且不小于maxX-binWidth
|
||
binSize -= 1;
|
||
}
|
||
vXBin.resize(xSize);
|
||
vYBin.resize(binSize);
|
||
for (int i = 0; i < binSize; ++i) vYBin[i] = 0;
|
||
for (int i = 0; i < xSize; ++i) {
|
||
int binIdx = (int)((x[i] + binWidth / 2) / binWidth);
|
||
if (binIdx >= binSize) binIdx = binSize - 1;
|
||
vYBin[binIdx] += 1;
|
||
}
|
||
// 按大小顺序将修改后的x数值存储在vXBin中，点的顺序不同，训练出的高斯混合模型参数会有一些不同。
|
||
int xIdx = 0;
|
||
for (int i = 0; i < binSize; ++i) {
|
||
for (int j = 0; j < vYBin[i]; ++j) {
|
||
vXBin[xIdx++] = i * binWidth;
|
||
}
|
||
}
|
||
}
|
||
|
||
/* 将标准高斯模型训练出的参数转换成自定义的系数, 并返回拟合后的Y值向量 */
|
||
struct cmpFunc {
|
||
bool operator()(const pair<double, double>& a, const pair<double, double>& b) { return a.first < b.first; }
|
||
};
|
||
void GMMToFactorEY(GMM& gmm, double binWidth, vector<double> &vYBin, vector<double>& vFactor, vector<double>& vEY) {
|
||
/* 需要调整曲线的权重，来拟合高斯曲线，而不是用概率密度 */
|
||
double zoomFactorSum = 0.0;
|
||
vEY.resize(vYBin.size());
|
||
int topM = (int)(vYBin.size() / 4);
|
||
if (topM < 1) topM = 1;
|
||
/* 用堆排序的方式取前topM个最大值, 用来计算缩放参数*/
|
||
priority_queue<pair<double, double>, vector<pair<double, double> >, cmpFunc> pqTopM;
|
||
|
||
for (int i = 0; i < vYBin.size(); ++i) {
|
||
double xVal = i * binWidth;
|
||
double probVal = gmm.GetProbability(&xVal);
|
||
vEY[i] = probVal;
|
||
pqTopM.push(make_pair(vYBin[i], probVal));
|
||
}
|
||
|
||
for (int i = 0; i < topM; ++i) {
|
||
pair<double, double> topEle = pqTopM.top();
|
||
pqTopM.pop();
|
||
zoomFactorSum += topEle.first / topEle.second;
|
||
}
|
||
|
||
double zoomFactor = zoomFactorSum / topM;
|
||
|
||
for (int i = 0; i < vEY.size(); ++i) {
|
||
vEY[i] *= zoomFactor;
|
||
}
|
||
vFactor.clear();
|
||
vFactor.push_back(zoomFactor * gmm.Prior(0) / sqrt(2 * M_PI * *gmm.Variance(0)));
|
||
vFactor.push_back(*gmm.Mean(0));
|
||
vFactor.push_back(sqrt(2 * *gmm.Variance(0)));
|
||
vFactor.push_back(zoomFactor * gmm.Prior(1) / sqrt(2 * M_PI * *gmm.Variance(1)));
|
||
vFactor.push_back(*gmm.Mean(1));
|
||
vFactor.push_back(sqrt(2 * *gmm.Variance(1)));
|
||
}
|
||
|
||
/* 计算平均数 */
|
||
template <typename T>
|
||
T Average(vector<T>& vVal) {
|
||
T sumVal = T(0);
|
||
for (int i = 0; i < vVal.size(); ++i) {
|
||
sumVal += vVal[i];
|
||
}
|
||
return sumVal / vVal.size();
|
||
}
|
||
|
||
/* 计算平方的均值 */
|
||
template <typename T>
|
||
T SquareAverage(vector<T>& vVal) {
|
||
vector<T> vSquare(vVal.size());
|
||
for (int i = 0; i < vVal.size(); ++i) {
|
||
vSquare[i] = vVal[i] * vVal[i];
|
||
}
|
||
return Average(vSquare);
|
||
}
|
||
|
||
/* 计算向量x和y的相关距离, 向量维度必须相等*/
|
||
double CorrelationDistance(vector<double>& vX, vector<double>& vY) {
|
||
vector<double> vXY(vX.size());
|
||
for (int i = 0; i < vXY.size(); ++i) {
|
||
vXY[i] = vX[i] * vY[i];
|
||
}
|
||
double uv = Average(vXY);
|
||
double uu = SquareAverage(vX);
|
||
double vv = SquareAverage(vY);
|
||
|
||
double dist = 1.0 - uv / sqrt(uu * vv);
|
||
return abs(dist);
|
||
}
|
||
|
||
/* 处理一个知识颗粒 */
|
||
struct ThreadParam {
|
||
fs::path matFilePath;
|
||
fs::path outFilePath;
|
||
};
|
||
void ThreadProcessData(const ThreadParam& param) {
|
||
const fs::path& matFilePath = param.matFilePath;
|
||
const fs::path& outFilePath = param.outFilePath;
|
||
double* hs = nullptr;
|
||
int rowNum = 0;
|
||
int colNum = 0;
|
||
|
||
hs = ReadMtxDouble(matFilePath.string(), "hs", &rowNum, &colNum);
|
||
|
||
vector<double>vXBin;
|
||
vector<double>vYBin;
|
||
vector<double>vEY;
|
||
vector<double>vFactor;
|
||
/* 用来保存数据，存入mat文件 */
|
||
vector<double>vDist(rowNum);
|
||
vector<double>vFactorAll;
|
||
for (int i = 0; i < rowNum; ++i) {
|
||
PutXtoBin(hs + i * colNum, colNum, 0.2, vXBin, vYBin);
|
||
|
||
GMM gmm(1, 2); // 1维， 2个高斯模型
|
||
gmm.Train(vXBin.data(), (int)vXBin.size());
|
||
|
||
GMMToFactorEY(gmm, 0.2, vYBin, vFactor, vEY);
|
||
vDist[i] = CorrelationDistance(vYBin, vEY);
|
||
vFactorAll.insert(vFactorAll.end(), vFactor.begin(), vFactor.end());
|
||
}
|
||
/* 写入matlab文件 */
|
||
MATFile* pMatFile = matOpen(outFilePath.string().c_str(), "w");
|
||
SaveMtxDouble(vFactorAll.data(), pMatFile, "factor", rowNum, 6);
|
||
SaveMtxDouble(vDist.data(), pMatFile, "correlation", rowNum, 1);
|
||
matClose(pMatFile);
|
||
|
||
delete[] hs;
|
||
}
|
||
|
||
/* 程序入口 */
|
||
int main(int argc, const char** argv) {
|
||
|
||
if (argc != 5) {
|
||
cerr << "This program should take 4 arguments(1.parrent Dir; 2. mat file suffix; 3. out mat filename; 4. thread number)!" << endl;
|
||
return 1;
|
||
}
|
||
string parrentDir(argv[1]); // 知识颗粒的父目录名称
|
||
string hsMatSuffix(argv[2]); // hs矩阵对应的mat文件的后缀名（可以是全文件名，可以是文件名后缀，必须保证唯一）
|
||
fs::path outFileName(argv[3]);
|
||
ThreadPool thPool(8);
|
||
clock_t begin, finish;
|
||
begin = clock();
|
||
|
||
/* 遍历所有的知识颗粒目录，逐一进行处理 */
|
||
for (auto& childDir : fs::directory_iterator(parrentDir)) {
|
||
fs::path outFilePath = childDir / outFileName;
|
||
for (auto& file : fs::directory_iterator(childDir)) {
|
||
const string& fileName = file.path().filename().string();
|
||
auto rPos = fileName.rfind(hsMatSuffix);
|
||
if (rPos != string::npos && fileName.size() - rPos == hsMatSuffix.size()) {
|
||
ThreadParam tParam = { file, outFilePath };
|
||
thPool.enqueue(ThreadProcessData, tParam);
|
||
}
|
||
}
|
||
}
|
||
thPool.~ThreadPool();
|
||
finish = clock();
|
||
cout << "GMM Total time: " << (double)(finish - begin) / CLOCKS_PER_SEC << " s" << endl;
|
||
return 0;
|
||
} |