NICE_VisLearning/features/simplefeatures/progs/progCodebookRandomForest.cpp

/**
  * @brief Extremely randomized clustering forest program for Matlab input data.
  *
  * @author Johannes Ruehle
  * @date 10/05/2014
  */

#ifdef NICE_USELIB_MEX
#ifdef NICE_USELIB_MATIO

#include <string>
#include <exception>
#include <iostream>
#include <fstream>

//----------
#include "vislearning/features/simplefeatures/CodebookRandomForest.h"
#include "vislearning/features/fpfeatures/VectorFeature.h"

#include "vislearning/cbaselib/FeaturePool.h"

#ifdef NICE_USELIB_MATIO
#include <core/matlabAccess/MatFileIO.h>

const bool verbose = false;
const bool verboseStartEnd = true;

using namespace OBJREC;
using namespace NICE;
using namespace std;

#undef DEBUG_VERBOSE

struct structCommands
{
    QString sFunction;
    QString sFileTrainData;
    QString sFileTrainDataLabels;
    QString sConfigFile;

    QString sFileStoreClassifier;   // txt file storing the config of the trained codebook rdf
    QString sFileStoreResult;       // matlab mat file storing the generated histogram

};

bool loadMatlabMatrix(const std::string &sFilename, const std::string &matrix_name, NICE::Matrix &p_Matrix)
{

    NICE::MatFileIO matlab_file(sFilename, MAT_ACC_RDONLY);

#ifdef DEBUG_VERBOSE
    // Show the number of variables in the file
    int vars_in_file = matlab_file.getNumberOfVariables();

    std::cout << vars_in_file << " Variables in " << sFilename << "\n";

    // Load the matrix
    std::cout << "Loading matrix \"" << matrix_name << "\"...\n";
#endif
    // Check if the variable is a matrix
    matvar_t* matrix_variable = matlab_file.getVariableViaName(matrix_name);
    if(matrix_variable == NULL)
    {
        std::cout << "variable is not found in mat file.\n";
        return false;
    }

    if(matrix_variable->rank != 2) {
        std::cout << "Variable is not a matrix. Rank: " << matrix_variable->rank << ".\n";
        return false;
    }

    // Read the dimensions
    int cols = matrix_variable->dims[1];
    int rows = matrix_variable->dims[0];
    std::cout << "Dimensions: " << cols << " x " << rows << "\n";

    // Read the matrix into a vector of vectors
    std::vector< std::vector<double> > matrix_vecvec(rows, std::vector<double>(cols));
    matlab_file.getFeatureMatrixViaName(matrix_vecvec, matrix_name);

    // Now, we want a NICE matrix
    //NICE::MatrixT<double> matrix(rows, cols);
    p_Matrix.resize(rows, cols);
    for(int i = 0; i < rows; i++) {
        for(int j = 0; j < cols; j++) {
            p_Matrix(i,j) = matrix_vecvec[i][j];
        }
    }

    return true;
}

NICE::Matrix* loadMatlabVec(const std::string &sFilename, const std::string &matrix_name)
{
    NICE::Matrix *pMatrix = NULL;

    NICE::MatFileIO *matFile = new NICE::MatFileIO(sFilename, MAT_ACC_RDONLY );

    matvar_t *t = matFile->getVariableViaName(matrix_name);
    if ( t->class_type == MAT_C_DOUBLE)
    {
        double *pD = (double*)( t->data );
        pMatrix = new NICE::Matrix(pD ,  (int)t->dims[0], (int)t->dims[1], Matrix::copy );
    }
    else
    {
        std::cerr << "raw format of matlab matrix not supported" << std::endl;
    }

    Mat_VarFree(t);
    delete matFile;

    return pMatrix;
}


bool saveMatlabVector(const std::string &sFilename, const NICE::Vector &p_Vector, int p_iFodID)
{
    std::ofstream ofs;
    ofs.open (sFilename.c_str(), std::ofstream::out);
    if (!ofs.is_open())
        return false;
    ofs << p_iFodID << " #fodID" << std::endl;
    ofs << p_Vector.size() << std::endl;
    for(int i=0; i<p_Vector.size(); i++)
        ofs << p_Vector[i] << std::endl;
    ofs.close();

    return true;
}


bool storeClassifier(const structCommands &p_Command, const OBJREC::CodebookRandomForest *p_pCodebookRandomForest)
{
    if( p_Command.sFileStoreClassifier.isEmpty() )
        return false;

    std::string t_sDestinationSave = p_Command.sFileStoreClassifier.toStdString();
    std::ofstream ofs;
    ofs.open (t_sDestinationSave.c_str(), std::ofstream::out);
    p_pCodebookRandomForest->store( ofs );
    ofs.close();

    return true;

}

bool restoreClassifier(const structCommands &p_Command, OBJREC::CodebookRandomForest *p_pCodebookRandomForest)
{
    if( p_Command.sFileStoreClassifier.isEmpty() )
        return false;

    if (p_pCodebookRandomForest == NULL )
        return false;

    std::string t_sDestinationSave = p_Command.sFileStoreClassifier.toStdString();
    std::ifstream ifs2;
    ifs2.open (t_sDestinationSave.c_str() );
    p_pCodebookRandomForest->restore( ifs2 );
    ifs2.close();

    return true;
}

bool createAndTrain( const structCommands &p_Command)
{
    if( p_Command.sConfigFile.isEmpty() )
    {
        std::cout << "no config file provided. Exiting" << std::endl;
        return false;
    }
    NICE::Config t_conf = NICE::Config( p_Command.sConfigFile.toStdString() );

    Matrix *t_pMatDataTrain       = loadMatlabVec( p_Command.sFileTrainData.toStdString(), "matFeatures");
    if( t_pMatDataTrain == NULL )
    {
        std::cout << "Training data Matrix couldn't be loaded" << std::endl;
        return 0;
    }
#ifdef DEBUG_VERBOSE
    for(int i = 0; i<10; i++)
    {
        std::cerr << (*t_pMatDataTrain)(i,0) << " ## " << (*t_pMatDataTrain)(0,i) << std::endl;
    }
#endif
    Matrix *t_pMatDataTrainLabels = loadMatlabVec( p_Command.sFileTrainDataLabels.toStdString(), "matLabels");
    if( t_pMatDataTrainLabels == NULL )
    {
        std::cout << "Training data label Matrix couldn't be loaded" << std::endl;
        return 0;
    }
    int iNumFeatureDimension = t_pMatDataTrain->rows();

    NICE::Vector t_vecLabelsTrain(t_pMatDataTrainLabels->getDataPointer(), t_pMatDataTrainLabels->rows(), Vector::external);

    OBJREC::Examples examplesTrain;

    bool bRet = OBJREC::Examples::wrapExamplesAroundFeatureMatrix( *t_pMatDataTrain, t_vecLabelsTrain, examplesTrain );
    if( !bRet )
    {
        std::cout << "createAndTrain: Error creating Examples from raw feature matrix and labels." << std::endl;
        return 0;
    }

    //----------------- create raw feature mapping -------------
    OBJREC::FeaturePool fp;
    OBJREC::VectorFeature *pVecFeature = new OBJREC::VectorFeature(iNumFeatureDimension);
    pVecFeature->explode(fp);

#ifdef DEBUG_VERBOSE
    //----------------- debug features -------------
    OBJREC::Example t_Exp = examplesTrain[0].second;
    NICE::Vector t_FeatVector;
    fp.calcFeatureVector(t_Exp, t_FeatVector);
    std::cerr << "first full Feature Vec: " <<t_FeatVector << std::endl;
#endif
    //----------------- train our random Forest -------------
    OBJREC::FPCRandomForests *pRandForest = new OBJREC::FPCRandomForests(&t_conf,"RandomForest");
    pRandForest->train(fp, examplesTrain);

    //----------------- create codebook ERC clusterer -------------
    int nMaxDepth        = t_conf.gI("CodebookRandomForest", "maxDepthTree",10);
    int nMaxCodebookSize = t_conf.gI("CodebookRandomForest", "maxCodebookSize",100);
#ifdef DEBUG_VERBOSE
    std::cerr << "maxDepthTree " << nMaxDepth << std::endl;
    std::cerr << "nMaxCodebookSize " << nMaxCodebookSize << std::endl;
#endif
    OBJREC::CodebookRandomForest *pCodebookRandomForest = new OBJREC::CodebookRandomForest(pRandForest, nMaxDepth,nMaxCodebookSize);


    //----------------- store classifier in file ---------------------
    bool bSuccess = storeClassifier(p_Command, pCodebookRandomForest);

    //----------------- clean up -------------

    delete pCodebookRandomForest;

    delete pVecFeature;
    pVecFeature = NULL;
    // delete all "exploded" features, they are internally cloned in the random trees anyway
    fp.destroy();
    //
    examplesTrain.clean();

    delete t_pMatDataTrain;
    delete t_pMatDataTrainLabels;

    return true;
}


bool generateHistogram( const structCommands &p_Command)
{

    Matrix *t_pMatFodID = loadMatlabVec( p_Command.sFileTrainData.toStdString(), "fodID");
    if( t_pMatFodID == NULL )
    {
        std::cout << "Data Matrix didn't include a fodID, so couldn't be loaded" << std::endl;
        return 0;
    }
    int iFodID = (*t_pMatFodID)(0,0);

    Matrix *t_pMatDataTrain       = loadMatlabVec( p_Command.sFileTrainData.toStdString(), "matFeatures");
    if( t_pMatDataTrain == NULL )
    {
        std::cout << "Data Matrix couldn't be loaded" << std::endl;
        return 0;
    }

    //----------------- restore trained codebook forest -------------
    OBJREC::CodebookRandomForest *pCodebookRandomForest = new OBJREC::CodebookRandomForest(-1,-1);
    if( !restoreClassifier(p_Command, pCodebookRandomForest ) )
    {
        std::cout << "Error restoring codebook random forest" << std::endl;
        return false;
    }

    size_t numTrainSamples      = t_pMatDataTrain->cols();
    size_t iNumFeatureDimension = t_pMatDataTrain->rows();
    size_t iNumCodewords        = pCodebookRandomForest->getCodebookSize();
#ifdef DEBUG_VERBOSE
    std::cerr << "numTrainSamples "      << numTrainSamples         << std::endl;
    std::cerr << "iNumFeatureDimension " << iNumFeatureDimension    << std::endl;
    std::cerr << "iNumCodewords "        << iNumCodewords           << std::endl;
#endif

    //----------------- parse config options  -------------
    bool bVerboseOutput = false;
//    if( nrhs > 3)
//    {
//        NICE::Config conf = parseParametersERC(prhs+3, nrhs-3 );
//        bVerboseOutput = conf.gB("CodebookRandomForest", "verbose", false);
//    }

    //----------------- quantize samples into histogram -------------
    NICE::Vector histogram(iNumCodewords, 0.0f);

    const double *pDataPtr = t_pMatDataTrain->getDataPointer();
    int t_iCodebookEntry; double t_fWeight; double t_fDistance;
    for (size_t i = 0; i < numTrainSamples; i++, pDataPtr+= iNumFeatureDimension )
    {
        const NICE::Vector t_VecTrainData( pDataPtr , iNumFeatureDimension);
        pCodebookRandomForest->voteVQ(t_VecTrainData, histogram, t_iCodebookEntry, t_fWeight, t_fDistance );
        if(bVerboseOutput)
            std::cerr << i << ": " << "CBEntry " << t_iCodebookEntry << " Weight: " << t_fWeight << " Distance: " << t_fDistance << std::endl;
    }

    // store histogram
    bool bSuccess = saveMatlabVector(p_Command.sFileStoreResult.toStdString(), histogram , iFodID);

    //----------------- clean up -------------

    delete pCodebookRandomForest;
    delete t_pMatDataTrain;

    return bSuccess;
}


#endif
int main(int argc, char **argv)
{

#ifdef NICE_USELIB_MATIO
#ifndef __clang__
#ifndef __llvm__
    std::set_terminate(__gnu_cxx::__verbose_terminate_handler);
#endif
#endif

    structCommands sCommand;

    QString sCmdArg;
    int iCurrArgIdx = 1;
    while(iCurrArgIdx < argc)
    {
        sCmdArg = QString(argv[iCurrArgIdx]);

        if    ( sCmdArg == "--function" )
        {
            iCurrArgIdx++;
            sCommand.sFunction = QString(argv[iCurrArgIdx]);
        }
        else if( sCmdArg == "--config" )
        {
            iCurrArgIdx++;
            sCommand.sConfigFile = QString(argv[iCurrArgIdx]);
        }

        else if( sCmdArg == "--traindata" )
        {
            iCurrArgIdx++;
            sCommand.sFileTrainData = QString(argv[iCurrArgIdx]);
        }
        else if( sCmdArg == "--traindatalabels" )
        {
            iCurrArgIdx++;
            sCommand.sFileTrainDataLabels = QString(argv[iCurrArgIdx]);
        }
        else if( sCmdArg == "--results" )
        {
            iCurrArgIdx++;
            sCommand.sFileStoreResult = QString(argv[iCurrArgIdx]);
        }
        else if( sCmdArg == "--classifier" )
        {
            iCurrArgIdx++;
            sCommand.sFileStoreClassifier = QString(argv[iCurrArgIdx]);
        }
        else if( sCmdArg == "--help" )
        {
//             print_usage();
             return 0;
        }
        else
        {
            std::cout << "unknown command arg: " << sCmdArg.toStdString() << std::endl;
        }

        iCurrArgIdx++;
    }

    ///////////////////////////////////////////////////
    try
    {
        if( sCommand.sFunction.compare("createAndTrain") == 0)
        {
            bool bSuccess = createAndTrain(sCommand);
        }
        else if( sCommand.sFunction.compare("generateHistogram") == 0)
        {
            bool bSuccess = generateHistogram(sCommand);
        }
    }
    catch(std::exception &e)
    {
        std::cerr << "exception occured: " << e.what() << std::endl;
    }
#else
    return -1;
#endif	

    return 0;
}
#endif //#ifdef NICE_USELIB_MATIO
#endif