main.cpp

#include <stdio.h>
#include <stdlib.h>
//#include <iostream>
#include <fstream>
#include <string.h>
#include <string>
#include <math.h>
#include <vector>
#include <deque>

#include "config.h"
#include "Data.h"
#include "Scheduler.h"
#include "StreamOutput.h"
#include "AlgorithmExploration.h"
#include "NumericalTools.h"

using namespace std;

StreamOutput cout;

int main ( int argc, char *argv[] )
{
    // number of arguments check
    if ( argc <= 1 )
    {
        cout<<"1) Please supply at least one argument [e] for algorithm exploration"<<endl;
        cout<<"2) Please supply a dataset directory and [t,p,b,x,y] for training, prediction, blending, bagging or boosting"<<endl;
        cout<<"   Example: ./EnsembleLearningFramework MNIST t"<<endl<<endl;
        exit ( 0 );
    }

    // set max. number of threads for OPENMP sections
    Framework::setMaxThreads ( omp_get_max_threads() );  // store max. number of threads of the machine

    // Algorithm exploration
    // one argument 'e'
    if ( argc == 2 )
        if ( argv[1][0] == 'e' || argv[1][0] == 'E' )
        {
            AlgorithmExploration a;
            a.start();
            exit ( 0 );
        }

    // training/prediction/blending
    // two arguments: 'DATASET'  and 't' 'p' or 'b' for train, predict or blend
    if ( argc <= 2 )
    {
        cout<<"Please supply a dataset directory and [t,p,b,x,y] for training, prediction, blending, bagging or boosting"<<endl;
        cout<<"Example: ./EnsembleLearningFramework MNIST t"<<endl<<endl;
        exit ( 0 );
    }
    if ( argc >= 4 )
        Framework::setAdditionalStartupParameter ( argv[3] );

    // the scheduler is responsible for sequential training of the algorithms
    // at the begining, the master description file is read
    Scheduler s;
    s.readMasterDscFile ( argv[1], "Master.dsc" );

    // training mode
    if ( argv[2][0] == 't' || argv[2][0] == 'T' )
        s.train();

    // prediction mode
    if ( argv[2][0] == 'p' || argv[2][0] == 'P' )
        s.predict();

    // blending mode (after successful training)
    if ( argv[2][0] == 'b' || argv[2][0] == 'B' )
        s.blend();

    // OBSOLETE! bagging: produce a set of predictions with slightly modified train set and average them
    if ( argv[2][0] == 'x' || argv[2][0] == 'X' )
        s.bagging();

    // OBSOLETE! boosting: reweight sample probabilites based on train error
    if ( argv[2][0] == 'y' || argv[2][0] == 'Y' )
        s.boosting();

    return 0;
}