432 lines
14 KiB
C++
432 lines
14 KiB
C++
#include <mex.h>
|
||
#include <mat.h>
|
||
#include <iostream>
|
||
#include <algorithm>
|
||
#include <string>
|
||
#include <unordered_set>
|
||
#include <ctime>
|
||
#include <vector>
|
||
#include <queue>
|
||
#include <memory>
|
||
#include <thread>
|
||
#include <mutex>
|
||
#include <condition_variable>
|
||
#include <future>
|
||
#include <functional>
|
||
#include <stdexcept>
|
||
#include <unordered_map>
|
||
#include <set>
|
||
#include <fstream>
|
||
#include <random>
|
||
#include <cmath>
|
||
#include <stdlib.h>
|
||
#include <limits.h>
|
||
#include <atomic>
|
||
using std::cout;
|
||
using std::endl;
|
||
using namespace std;
|
||
|
||
#define STRING_BUF_SIZE 204800
|
||
|
||
/* 读取二层cell包裹的字符串,和数值，ds,fr */
|
||
#define OUTER_FOR_BEGIN \
|
||
rowNum = (int)mxGetM(pMxArray); \
|
||
colNum = (int)mxGetN(pMxArray); \
|
||
for (int i = 0; i < rowNum; ++i) { \
|
||
for (int j = 0; j < colNum; ++j) { \
|
||
mxArray* pCell = mxGetCell(pMxArray, j * rowNum + i); \
|
||
int childRowNum = (int)mxGetM(pCell); \
|
||
int childColNum = (int)mxGetN(pCell);
|
||
|
||
#define OUTER_FOR_END \
|
||
} \
|
||
}
|
||
|
||
#define INNTER_FOR_BEGIN \
|
||
for (int ii = 0; ii < childRowNum; ii++) { \
|
||
for (int jj = 0; jj < childColNum; jj++) { \
|
||
mxArray *pChildCell = mxGetCell(pCell, jj * childRowNum + ii);
|
||
#define INNTER_FOR_END \
|
||
} \
|
||
}
|
||
// 将matlab存储方式转换成c存储方式
|
||
#define TRANS_ROW_COL(dst, src, rowNum, colNum) \
|
||
for (int rowI = 0; rowI < rowNum; ++rowI) { \
|
||
for (int colJ = 0; colJ < colNum; ++colJ) { \
|
||
dst[rowI * colNum + colJ] = src[colJ * rowNum + rowI]; \
|
||
} \
|
||
}
|
||
|
||
class ThreadPool {
|
||
public:
|
||
ThreadPool(size_t);
|
||
template<class F, class... Args>
|
||
auto enqueue(F&& f, Args&&... args)
|
||
->std::future<typename std::result_of<F(Args...)>::type>;
|
||
~ThreadPool();
|
||
private:
|
||
// need to keep track of threads so we can join them
|
||
std::vector< std::thread > workers;
|
||
// the task queue
|
||
std::queue< std::function<void()> > tasks;
|
||
|
||
// synchronization
|
||
std::mutex queue_mutex;
|
||
std::condition_variable condition;
|
||
bool stop;
|
||
};
|
||
|
||
// the constructor just launches some amount of workers
|
||
inline ThreadPool::ThreadPool(size_t threads)
|
||
: stop(false)
|
||
{
|
||
for (size_t i = 0;i < threads;++i)
|
||
workers.emplace_back(
|
||
[this]
|
||
{
|
||
for (;;)
|
||
{
|
||
std::function<void()> task;
|
||
|
||
{
|
||
std::unique_lock<std::mutex> lock(this->queue_mutex);
|
||
this->condition.wait(lock,
|
||
[this] { return this->stop || !this->tasks.empty(); });
|
||
if (this->stop && this->tasks.empty())
|
||
return;
|
||
task = std::move(this->tasks.front());
|
||
this->tasks.pop();
|
||
}
|
||
|
||
task();
|
||
}
|
||
}
|
||
);
|
||
}
|
||
|
||
// add new work item to the pool
|
||
template<class F, class... Args>
|
||
auto ThreadPool::enqueue(F && f, Args&&... args)
|
||
-> std::future<typename std::result_of<F(Args...)>::type>
|
||
{
|
||
using return_type = typename std::result_of<F(Args...)>::type;
|
||
|
||
auto task = std::make_shared< std::packaged_task<return_type()> >(
|
||
std::bind(std::forward<F>(f), std::forward<Args>(args)...)
|
||
);
|
||
|
||
std::future<return_type> res = task->get_future();
|
||
{
|
||
std::unique_lock<std::mutex> lock(queue_mutex);
|
||
|
||
// don't allow enqueueing after stopping the pool
|
||
if (stop)
|
||
throw std::runtime_error("enqueue on stopped ThreadPool");
|
||
|
||
tasks.emplace([task]() { (*task)(); });
|
||
}
|
||
condition.notify_one();
|
||
return res;
|
||
}
|
||
|
||
// the destructor joins all threads
|
||
inline ThreadPool::~ThreadPool()
|
||
{
|
||
{
|
||
std::unique_lock<std::mutex> lock(queue_mutex);
|
||
stop = true;
|
||
}
|
||
condition.notify_all();
|
||
for (std::thread& worker : workers)
|
||
worker.join();
|
||
}
|
||
|
||
// 将二维索引转成一维的索引
|
||
inline int Get1DIndex(int colNum, int row, int col) {
|
||
return row * colNum + col;
|
||
}
|
||
|
||
// 读取G结构体中的ds和fr
|
||
void GetFrDs(const mxArray* pMxParent, vector<vector<string> >& vvDs, vector<vector<double> >& vvFr) {
|
||
// 读取ds字符串
|
||
int rowNum, colNum;
|
||
char *strBuf = new char[STRING_BUF_SIZE];
|
||
mxArray* pMxArray = mxGetField(pMxParent, 0, "ds"); // ds
|
||
OUTER_FOR_BEGIN
|
||
vvDs.push_back(vector<string>());
|
||
vvDs.back().resize(childRowNum * childColNum);
|
||
INNTER_FOR_BEGIN
|
||
if (mxGetString(pChildCell, strBuf, STRING_BUF_SIZE) != 0) {
|
||
cout << "String is too large to fit in the buffer! " << i + 1 << '\t' << j + 1 << endl;
|
||
delete[]strBuf;
|
||
return;
|
||
}
|
||
vvDs.back()[ii * childColNum + jj] = strBuf;
|
||
auto& lastStr = vvDs.back()[ii * childColNum + jj];
|
||
transform(lastStr.begin(), lastStr.end(), lastStr.begin(), ::toupper); // 转成大写
|
||
INNTER_FOR_END
|
||
OUTER_FOR_END
|
||
|
||
// 读取fr数值
|
||
pMxArray = mxGetField(pMxParent, 0, "fr"); // fr
|
||
OUTER_FOR_BEGIN
|
||
vvFr.push_back(vector<double>());
|
||
vvFr.back().resize(childRowNum * childColNum);
|
||
double* pVal = (double*)mxGetData(pCell); //获取指针
|
||
TRANS_ROW_COL(vvFr.back(), pVal, childRowNum, childColNum); // 行列存储方式转换
|
||
OUTER_FOR_END
|
||
delete[]strBuf;
|
||
}
|
||
|
||
/* 读取abs */
|
||
void GetAbstract(const mxArray* pMxAbs, vector<string>& vAbs) {
|
||
int rowNum = (int)mxGetM(pMxAbs);
|
||
int colNum = (int)mxGetN(pMxAbs);
|
||
char *strBuf = new char[STRING_BUF_SIZE];
|
||
|
||
vAbs.resize(rowNum * colNum);
|
||
for (int i = 0; i < rowNum; ++i) {
|
||
for (int j = 0; j < colNum; ++j) {
|
||
mxArray* pCell = mxGetCell(pMxAbs, j * rowNum + i);
|
||
if (mxGetString(pCell, strBuf, STRING_BUF_SIZE) != 0) {
|
||
cout << "String is too large to fit in the buffer! " << i + 1 << '\t' << j + 1 << endl;
|
||
delete[]strBuf;
|
||
return;
|
||
}
|
||
vAbs[i * colNum + j] = strBuf;
|
||
}
|
||
}
|
||
delete[]strBuf;
|
||
}
|
||
|
||
// 保存由一维cell组成的字符串数组
|
||
mxArray* writeToMatString1DCell(vector<string>& vStr) {
|
||
mxArray* pCellMtx = mxCreateCellMatrix(1, vStr.size());
|
||
for (int j = 0; j < vStr.size(); ++j) {
|
||
mxArray* mxStr = mxCreateString(vStr[j].c_str());
|
||
mxSetCell(pCellMtx, j, mxStr);
|
||
}
|
||
return pCellMtx;
|
||
}
|
||
|
||
// 保存由二维cell组成的字符串数组
|
||
mxArray* writeToMatString2DCell(vector<vector<string>>& vvStr) {
|
||
mxArray* pCellMtx = mxCreateCellMatrix(1, vvStr.size());
|
||
for (int i = 0; i < vvStr.size(); ++i) {
|
||
mxArray* pChildCellMtx = writeToMatString1DCell(vvStr[i]);
|
||
mxSetCell(pCellMtx, i, pChildCellMtx);
|
||
}
|
||
return pCellMtx;
|
||
}
|
||
|
||
// 将结果写入mxArray, 作为后续的返回值
|
||
mxArray* writeToMatDouble(const double* data, int rowNum, int colNum) {
|
||
mxArray* pWriteArray = NULL;//matlab格式矩阵
|
||
int len = rowNum * colNum;
|
||
//创建一个rowNum*colNum的矩阵
|
||
pWriteArray = mxCreateDoubleMatrix(rowNum, colNum, mxREAL);
|
||
//把data的值赋给pWriteArray指针
|
||
memcpy((void*)(mxGetPr(pWriteArray)), (void*)data, sizeof(double) * len);
|
||
return pWriteArray; // 赋值给返回值
|
||
}
|
||
|
||
/* 多线程计算信息熵 */
|
||
struct TPEntropy {
|
||
vector<string>* pvDs;
|
||
vector<double>* pvFr;
|
||
vector<unordered_set<string>>* pvusAbsWord;
|
||
double* pHs;
|
||
};
|
||
|
||
void ThreadCalcEntropy(TPEntropy& param) {
|
||
vector<string>& vDs = *param.pvDs; // 这一组ds
|
||
vector<double>& vFr = *param.pvFr; // frequency
|
||
vector<unordered_set<string>>& vusAbsWord = *param.pvusAbsWord;
|
||
double* hs = param.pHs;
|
||
const int numAbs = vusAbsWord.size();
|
||
const int numDsWord = vDs.size(); // 这一组数据中包含的单词数量
|
||
|
||
vector<vector<int> > vX(numAbs, vector<int>(numDsWord, 0));
|
||
// 检查知识颗粒中的词语是否出现在pubmed摘要的词语中
|
||
for (int i = 0; i < numAbs; ++i) {
|
||
for (int j = 0; j < numDsWord; ++j) {
|
||
if (vusAbsWord[i].find(vDs[j]) != vusAbsWord[i].end()) { // 这一组单词中的j索引位置的单词在第i个文献中出现过
|
||
vX[i][j] = 1;
|
||
}
|
||
}
|
||
}
|
||
|
||
// 找词汇的最高频率
|
||
double maxFr = *max_element(vFr.begin(), vFr.end());
|
||
// 将fr的数值规范化到（0，0.368）之间
|
||
const double normalMax = 0.368;
|
||
for (auto& frVal : vFr) frVal = frVal * normalMax / maxFr;
|
||
maxFr = normalMax;
|
||
// 对每个知识颗粒每一组数据，计算信息熵
|
||
for (int i = 0; i < numAbs; ++i) {
|
||
for (int j = 0; j < numDsWord; ++j) {
|
||
if (vX[i][j] == 1) {
|
||
hs[i] -= vFr[j] * log2(vFr[j]);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
/*
|
||
输入：
|
||
1. abs: 待感知的文献的摘要信息。
|
||
2. G: 知识颗粒，包含该程序需要的热词ds以及对应的频率fr。
|
||
输出：
|
||
1. hs: 信息熵，二维[len(知识颗粒)][len(文献)]
|
||
*/
|
||
void mexFunction(int nlhs, mxArray* plhs[], int nrhs, const mxArray* prhs[]) {
|
||
if (nrhs < 2) {
|
||
cout << "At least 2 arguments should be given for this function!" << endl;
|
||
return;
|
||
}
|
||
clock_t begin = clock(), mid, finish;
|
||
vector<string> vAbstract; // 读取abs1, 然后分割成一个一个的单词
|
||
GetAbstract(prhs[0], vAbstract);
|
||
|
||
vector<vector<string>> vvDs; // 每个知识颗粒的ds矩阵（词汇矩阵）
|
||
vector<vector<double>> vvFr; // 词汇对应的频率
|
||
GetFrDs(prhs[1], vvDs, vvFr);
|
||
|
||
int numThread = 1; // 是否打印信息, 1打印简单信息，2打印详细信息
|
||
if (nrhs > 2) {
|
||
double* pData = (double*)mxGetData(prhs[2]);
|
||
numThread = (int)pData[0];
|
||
if (numThread < 1) numThread = 1;
|
||
}
|
||
|
||
int flagPrint = 0; // 是否打印信息, 1打印简单信息，2打印详细信息
|
||
if (nrhs > 3) {
|
||
double* pData = (double*)mxGetData(prhs[3]);
|
||
flagPrint = (int)pData[0];
|
||
}
|
||
finish = clock();
|
||
if (flagPrint == 2) cout << "Load data time: " << (double)(finish - begin) / CLOCKS_PER_SEC << " s" << endl;
|
||
/* 将摘要信息分割成一个一个的词汇 */
|
||
mid = clock();
|
||
unordered_set<char> usWordChars; // 能组成单词的字符，要不要考虑数字？原版matlab是提取了数字的
|
||
for (int i = 65; i <= 90; i++) usWordChars.insert(char(i)); // A - Z
|
||
for (int i = 97; i <= 122; i++) usWordChars.insert(char(i)); // a - z
|
||
for (int i = 48; i <= 57; i++) usWordChars.insert(char(i)); // 0 - 9
|
||
usWordChars.insert('/'); usWordChars.insert('+'); usWordChars.insert('-');
|
||
vector<vector<string> > vvWordMtx(vAbstract.size()); // 初始大小为文章的个数
|
||
vector<unordered_set<string> > vusAbsWord(vAbstract.size()); // 将每篇文章摘要的单词放入hash表
|
||
for (int i = 0; i < vAbstract.size(); i++) {
|
||
auto& strAbs = vAbstract[i];
|
||
// 遍历摘要字符串的每一个字符，取出每一个单词
|
||
vector<string>& vWord = vvWordMtx[i];
|
||
if (strAbs.size() == 0) continue; // 摘要信息为空，跳过（一般不会出现这个情况）
|
||
int wordStartPos = 0;
|
||
while (wordStartPos < strAbs.size() && usWordChars.find(strAbs[wordStartPos]) == usWordChars.end())
|
||
wordStartPos++;
|
||
for (int curPos = wordStartPos + 1; curPos < strAbs.size(); ++curPos) {
|
||
if (usWordChars.find(strAbs[curPos]) == usWordChars.end()) { // 找到了分割符
|
||
vWord.push_back(strAbs.substr(wordStartPos, curPos - wordStartPos));
|
||
wordStartPos = curPos + 1; // 找下一个词语起始位置
|
||
while (wordStartPos < strAbs.size() && usWordChars.find(strAbs[wordStartPos]) == usWordChars.end())
|
||
wordStartPos++;
|
||
curPos = wordStartPos; // 循环会自动加1
|
||
}
|
||
}
|
||
// 将处理摘要之后的每个词语放入hash表
|
||
for (auto& word : vWord) {
|
||
string upWord(word);
|
||
transform(upWord.begin(), upWord.end(), upWord.begin(), ::toupper);
|
||
vusAbsWord[i].insert(upWord);
|
||
}
|
||
}
|
||
finish = clock();
|
||
if (flagPrint == 2) cout << "Split abstract time: " << (double)(finish - mid) / CLOCKS_PER_SEC << " s" << endl;
|
||
|
||
// 存放结果，用一维数组存放二维数据
|
||
mid = clock();
|
||
vector<double> hs;
|
||
// vector<double> hr;
|
||
const int numLiterature = vusAbsWord.size(); // pubmed 文件中包含的文献数量
|
||
const int numGroup = vvDs.size(); // ds包含的组数
|
||
hs.resize(numGroup * numLiterature);
|
||
// hr.resize(numLiterature * numGroup);
|
||
// 并行, 没有计算hr
|
||
ThreadPool thPool(numThread);
|
||
for (int groupIdx = 0; groupIdx < numGroup; ++groupIdx) {
|
||
TPEntropy tp = { &vvDs[groupIdx], &vvFr[groupIdx], &vusAbsWord, &hs[groupIdx * numLiterature] };
|
||
thPool.enqueue(ThreadCalcEntropy, tp);
|
||
}
|
||
thPool.~ThreadPool();
|
||
|
||
// // 串行
|
||
// for (int groupIdx = 0; groupIdx < numGroup; ++groupIdx) { // 遍历知识颗粒中的每一组
|
||
// vector<string>& vDs = vvDs[groupIdx]; // 这一组ds
|
||
// vector<double>& vFr = vvFr[groupIdx]; // frequency
|
||
// const int numWord = vDs.size(); // 这一组数据中包含的单词数量
|
||
// vector<vector<int> > vX(numLiterature, vector<int>(numWord, 0));
|
||
// // 检查知识颗粒中的词语是否出现在pubmed摘要的词语中
|
||
// for (int i = 0; i < numLiterature; ++i) {
|
||
// for (int j = 0; j < numWord; ++j) {
|
||
// if (vusAbsWord[i].find(vDs[j]) != vusAbsWord[i].end()) { // 这一组单词中的j索引位置的单词在第i个文献中出现过
|
||
// vX[i][j] = 1;
|
||
// }
|
||
// }
|
||
// }
|
||
//
|
||
// // 找词汇的最高频率
|
||
// double maxFr = *max_element(vFr.begin(), vFr.end());
|
||
// // 将fr的数值规范化到（0，0.368）之间
|
||
// const double normalMax = 0.368;
|
||
// for (auto& frVal : vFr) frVal = frVal * normalMax / maxFr;
|
||
// maxFr = normalMax;
|
||
// // 对每个知识颗粒每一组数据，计算信息熵
|
||
// for (int i = 0; i < numLiterature; ++i) {
|
||
// for (int j = 0; j < numWord; ++j) {
|
||
// if (vX[i][j] == 1) {
|
||
// hs[Get1DIndex(numLiterature, groupIdx, i)] -= vFr[j] * log2(vFr[j]);
|
||
// }
|
||
// }
|
||
// }
|
||
//
|
||
// // 找最高频词汇所在的索引位置
|
||
// vector<int> vMaxPos;
|
||
// int idx = 0;
|
||
// for_each(vFr.begin(), vFr.end(), [&idx, maxFr, &vMaxPos](double val) {
|
||
// if (val == maxFr) vMaxPos.push_back(idx);
|
||
// idx++;
|
||
// });
|
||
//
|
||
// for (int i = 0; i < numLiterature; ++i) {
|
||
// int cumulateX = 0; // 计算在最高频词汇处，x值的累加结果
|
||
// for (int j = 0; j < vMaxPos.size(); ++j) cumulateX += vX[i][vMaxPos[j]];
|
||
// if (cumulateX == vMaxPos.size()) { // 如果频率最高的词汇都出现在了文献中
|
||
// hr[Get1DIndex(numGroup, i, groupIdx)] = 1; // 应该是表示知识颗粒的这一组数据跟这篇文献相关性比较高
|
||
// }
|
||
// }
|
||
// }
|
||
finish = clock();
|
||
if (flagPrint == 2) cout << "Calc entropy time: " << (double)(finish - mid) / CLOCKS_PER_SEC << " s" << endl;
|
||
|
||
/* 将结果写入返回值 */
|
||
mid = clock();
|
||
if (nlhs > 0) {
|
||
int datasize = numGroup * numLiterature;
|
||
vector<double> vData(datasize);
|
||
for (int i = 0; i < numGroup; i++) for (int j = 0; j < numLiterature; j++)
|
||
vData[j * numGroup + i] = hs[i * numLiterature + j];
|
||
plhs[0] = writeToMatDouble(vData.data(), numGroup, numLiterature);
|
||
}
|
||
if (nlhs > 1) { // 将ws写入结果
|
||
plhs[1] = writeToMatString2DCell(vvWordMtx);
|
||
}
|
||
finish = clock();
|
||
if (flagPrint == 2) cout << "Write back data time: " << (double)(finish - mid) / CLOCKS_PER_SEC << " s" << endl;
|
||
|
||
finish = clock();
|
||
if(flagPrint) cout << "CalcEntropy Total time: " << (double)(finish - begin) / CLOCKS_PER_SEC << " s" << endl;
|
||
}
|
||
|
||
/* 供main调试调用 */
|
||
void mexFunctionWrap(int nlhs, mxArray* plhs[], int nrhs, const mxArray* prhs[]) {
|
||
mexFunction(nlhs, plhs, nrhs, prhs);
|
||
} |