MaxLogLikelihood.cpp

Go to the documentation of this file.

#include <shark/Data/Csv.h> //load the csv file
#include <shark/Algorithms/Trainers/NormalizeComponentsUnitVariance.h> //normalize

#include <shark/Algorithms/Trainers/NormalizeComponentsUnitVariance.h> //normalize

#include<shark/Models/RBFLayer.h>
#include<shark/Models/ConvexCombination.h>
#include<shark/Models/ConcatenatedModel.h>

#include<shark/ObjectiveFunctions/NegativeLogLikelihood.h>
#include<shark/Algorithms/GradientDescent/Rprop.h>
#include <shark/Algorithms/KMeans.h>

using namespace shark;
using namespace std;


int main(int argc, char **argv) {
    if(argc < 2) {
        cerr << "usage: " << argv[0] << " (filename)" << endl;
        return 1;
    }
    UnlabeledData<RealVector> data;
    try {
        importCSV(data, argv[1], ' ');
        Normalizer<> normalizer;
        NormalizeComponentsUnitVariance<> normalizingTrainer(true); // zero-mean
        normalizingTrainer.train(normalizer, data);
        data = normalizer(data);
    } 
    catch (...) {
        cerr << "unable to read data from file " <<  argv[1] << endl;
        return 1;
    }
    
    std::size_t inputs=dataDimension(data);
    std::size_t hiddens = 2;
    unsigned numberOfSteps = 100;

    //create mixture of gaussian mixture and initialize weights
    RBFLayer gaussians;
    gaussians.setStructure(inputs,hiddens);
    ConvexCombination combination;
    combination.setStructure(hiddens);
    ConcatenatedModel<RealVector,RealVector> mixture= gaussians >> combination;
    initRandomUniform(mixture,-0.1,0.1);
    kMeans(data, gaussians);
    
    //create error function
    NegativeLogLikelihood error(data,&mixture);
    //initialize Rprop
    IRpropPlus optimizer;
    error.init();
    optimizer.init(error);
    
    RealMatrix initCenters = gaussians.centers();
    RealVector initWidth = gaussians.gamma();
    
    std::cout<<0<<" "<<optimizer.solution().value<<std::endl;;
    for(unsigned step = 0; step != numberOfSteps; ++step){
        optimizer.step(error);
        if(step % 10 == 9)
            std::cout<<step<<" "<<optimizer.solution().value<<std::endl;;
    }
    
    std::cout<<"center solutions"<<std::endl;
    std::cout<<"init:"<<initCenters<<std::endl;
    std::cout<<"after optimization:"<<gaussians.centers()<<std::endl;
    std::cout<<"width solutions"<<std::endl;
    std::cout<<"init:"<<initWidth<<std::endl;
    std::cout<<"after optimization:"<<gaussians.gamma()<<std::endl;
}