Files
HandyGames/workspace/usingpca.h
2016-01-10 14:43:47 +01:00

314 lines
7.6 KiB
C++

#ifndef USINGPCA_H
#define USINGPCA_H
#include "pca/TrainPCA.h"
#include "pca/KNN.h"
#include "pca/aKNN.h"
std::vector<std::string> COLORS = {"#000000", "#0000ff", "#00ff00", "#ff0000", "#00ffff"};
struct Plot {
K::Gnuplot gp;
K::GnuplotSplot splot;
K::GnuplotSplotElementLines lines[5];
public:
Plot() {
for (int i = 0; i < 5; ++i) {lines[i].setColorHex(COLORS[i]);}
for (int i = 0; i < 5; ++i) {splot.add(&lines[i]);}
}
void add(int idx, std::vector<float>& vec) {
K::GnuplotPoint3 p3(vec[0], vec[1], vec[2]);
lines[idx].add(p3);
}
void clear() {
for (int i = 0; i < 5; ++i) {lines[i].clear();}
}
void show() {
gp.setDebugOutput(false);
gp.draw(splot);
gp.flush();
}
};
std::string getClass(const std::vector<ClassifiedFeature>& nns) {
std::unordered_map<std::string, int> map;
for(const ClassifiedFeature& nn : nns) { map[nn.className] += 1; }
for (auto& it : map) {
if (it.second > nns.size() * 0.75) {return it.first;}
}
return "";
}
struct ClassStats {
int counts[6] = {};
};
struct Stats{
int match;
int error;
int unknown;
Stats() : match(0), error(0), unknown(0) {;}
float getSum() {return match+error+unknown;}
};
std::vector<ClassifiedPattern> removePatterns(const std::vector<ClassifiedPattern>& patAll, const std::string& fileName) {
std::vector<ClassifiedPattern> res;
for (const ClassifiedPattern& pat : patAll) {
if (pat.belongsToFile(fileName)) {
continue;
} else {
res.push_back(pat);
}
}
return res;
}
template <int numFeatures> struct PCA {
aKNN<ClassifiedFeature, numFeatures> knn;
TrainPCA::Matrices m;
};
void runPCA() {
const int numFeatures = 10;
TrainPCA::Settings setTrain;
TrainPCA::Settings setClass; setClass.regionStart_ms += 25;
Plot p;
// convert all provided datasets into patterns
std::vector<ClassifiedPattern> srcTrain = TrainPCA::getAllData(setTrain);
std::vector<ClassifiedPattern> srcClass = TrainPCA::getAllData(setClass);
std::cout << "windows: " << srcTrain.size() << std::endl;
// error calculation
std::unordered_map<std::string, Stats> stats;
std::unordered_map<std::string, ClassStats> classStats;
int xx = 0;
std::unordered_map<std::string, PCA<numFeatures>*> pcas;
// try to classify each pattern
for (const ClassifiedPattern& patClassify : srcClass) {
// construct knn search for this leave-one-out ONLY ONCE
if (pcas.find(patClassify.fileName) == pcas.end()) {
std::cout << "constructing PCA for all files but " << patClassify.fileName << std::endl;
// remove all training patterns belonging to the same source file as the to be classifed pattern
std::vector<ClassifiedPattern> srcTrainLOO = removePatterns(srcTrain, patClassify.fileName);
// sanity check (have we removed all patterns?)
int diff = srcTrain.size() - srcTrainLOO.size();
if (diff < 200) {throw 1;}
p.clear();
PCA<numFeatures>* pca = new PCA<numFeatures>();
pcas[patClassify.fileName] = pca;
// train PCA using all pattern without those belonging to the same source file as the to-be-classified one
pca->m = TrainPCA::getMatrices(srcTrainLOO, numFeatures);
// calculate features and add them to the KNN
for (const ClassifiedPattern& pat : srcTrainLOO) {
K::DynColVector<float> vec = pca->m.A1 * K::PCAHelper<float>::toVector(pat.pattern);
std::vector<float> arr;
for (int i = 0; i < numFeatures; ++i) {arr.push_back(vec(i));}
pca->knn.add(ClassifiedFeature(pat.className, arr));
const int idx = Settings::classToInt(pat.className);
p.add(idx, arr);
}
pca->knn.build();
if (xx == 0) {
++xx;
std::ofstream out("/tmp/pca.gp"); p.gp.draw(p.splot); out << p.gp.getBuffer(); out.close();
}
//p.show();
//sleep(100);
}
{
PCA<numFeatures>* pca = pcas[patClassify.fileName];
// calculate features for the to-be-classified pattern
//const int idx = Settings::classToInt(pat.className);
K::DynColVector<float> vec = pca->m.A1 * K::PCAHelper<float>::toVector(patClassify.pattern);
// get KNN's answer
std::vector<float> arr;
for (int i = 0; i < numFeatures; ++i) {arr.push_back(vec(i));}
std::vector<ClassifiedFeature> neighbors = pca->knn.get(arr.data(), 5);
std::string gotClass = getClass(neighbors);
if (patClassify.className == gotClass) {stats["all"].match++; stats[patClassify.fileName].match++; stats[patClassify.className].match++;}
else if (gotClass == "") {stats["all"].unknown++; stats[patClassify.fileName].unknown++; stats[patClassify.className].unknown++;}
else {stats["all"].error++; stats[patClassify.fileName].error++; stats[patClassify.className].error++;}
int gotIdx = (gotClass == "") ? (5) : Settings::classToInt(gotClass);
++classStats[patClassify.className].counts[gotIdx];
}
}
for (auto& it : stats) {
Stats& stats = it.second;
std::cout << "'" <<it.first << "',";
std::cout << stats.match/stats.getSum() << ",";
std::cout << stats.error/stats.getSum() << ",";
std::cout << stats.unknown/stats.getSum();
std::cout << std::endl;
}
for (auto& it : classStats) {
ClassStats& stats = it.second;
std::cout << "'" << it.first << "',";
for (int i = 0; i < 6; ++i) {
std::cout << stats.counts[i] << ",";
}
std::cout << std::endl;
}
}
/*
#include <vector>
#include "sensors/SensorReader.h"
#include "Interpolator.h"
#include <eigen3/Eigen/Dense>
enum class PracticeType {
//REST,
JUMPING_JACK,
SITUPS,
PUSHUPS,
KNEEBEND,
FORWARDBEND,
};
struct Practice {
PracticeType type;
Recording rec;
std::vector<uint64_t> keyGyro;
//Practice(const PracticeType p, const Recording& rec, const std::vector<uint64_t>& keyGyro) : p(p), rec(rec), keyGyro(keyGyro) {;}
K::Interpolator<uint64_t, SensorGyro> getInterpol() const {
K::Interpolator<uint64_t, SensorGyro> interpol;
for (auto it : rec.gyro.values) {interpol.add(it.ts, it.val);}
interpol.makeRelative();
return interpol;
}
};
class UsingPCA {
public:
static Eigen::VectorXf getWindow(Practice& p, uint64_t pos) {
K::Interpolator<uint64_t, SensorGyro> interpol = p.getInterpol();
Eigen::VectorXf vec(600/50*3, 1);
int idx = 0;
for (int offset = -300; offset < 300; offset += 50) {
SensorGyro gyro = interpol.get(pos + offset);
vec(idx++,0) = (gyro.x);
vec(idx++,0) = (gyro.y);
vec(idx++,0) = (gyro.z);
}
std::cout << vec << std::endl;
return vec;
}
static std::vector<Eigen::VectorXf> getClassWindows(Practice& p) {
std::vector<Eigen::VectorXf> windows;
for (uint64_t pos = 1000; pos < 5000; pos += 500) {
Eigen::VectorXf window = getWindow(p, pos);
windows.push_back(window);
}
return windows;
}
static Eigen::MatrixXf getR(std::vector<Eigen::VectorXf>& vecs) {
Eigen::MatrixXf mat = Eigen::MatrixXf::Zero(vecs[0].rows(), vecs[0].rows());
for (const Eigen::VectorXf& vec : vecs) {
mat += vec * vec.transpose();
}
mat /= vecs.size();
return mat;
}
static Eigen::VectorXf getM(std::vector<Eigen::VectorXf>& vecs) {
Eigen::MatrixXf mat = Eigen::MatrixXf::Zero(vecs[0].rows(), vecs[0].cols());
for (const Eigen::VectorXf& vec : vecs) {
mat += vec;
}
mat /= vecs.size();
return mat;
}
static void run() {
std::vector<Practice> practices;
practices.push_back(
Practice {
PracticeType::JUMPING_JACK,
SensorReader::read("/mnt/firma/kunden/HandyGames/daten/jumpingjack/jumpingjack_gl_5_subject_3_left.txt"),
{1950, 2900, 3850, 4850, 5850, 6850, 7850, 8850, 9800, 10800, 11850}
}
);
std::vector<Eigen::VectorXf> windows = getClassWindows(practices.back());
Eigen::MatrixXf R = getR(windows);
Eigen::MatrixXf m = getM(windows);
Eigen::MatrixXf Q = R - (m * m.transpose());
Eigen::SelfAdjointEigenSolver<Eigen::MatrixXf> es;
es.compute(Q);
int i = 0;
}
};
*/
#endif // USINGPCA_H