regression.cc

Go to the documentation of this file.

/*

Copyright (c) 2003, Cornell University
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

   - Redistributions of source code must retain the above copyright notice,
       this list of conditions and the following disclaimer.
   - Redistributions in binary form must reproduce the above copyright
       notice, this list of conditions and the following disclaimer in the
       documentation and/or other materials provided with the distribution.
   - Neither the name of Cornell University nor the names of its
       contributors may be used to endorse or promote products derived from
       this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
THE POSSIBILITY OF SUCH DAMAGE.

*/

#include "machine.h"
#include "regressiontree.h"
#include "binaryobliquesplitter.h"
#include "probabilisticregressiontree.h"
#include "binaryobliqueprobabilisticsplitter.h"
#include "multinormal.h"
#include "skinymultinormal.h"
#include "linearregressor.h"
#include "streamdctrain.h"
#include <iostream>
#include "syncobjlist.h"
#include "fileprod.h"
#include "rpms.h"
#include <stdlib.h>
#include <time.h>
#include "gridinput.h"
#include "filecons.h"
#include <getopt.h>
#include <gsl/gsl_errno.h>

using namespace CLUS;

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

    char trainfilename[256];
    char prunefilename[256];
    int c_split = 0;
    int r_dim = 0;
    char modelfile[256];
    char testfilename[256];
    int griddim = 0;
    char rezfile[256];

    /// 0 means whole tree
    int MaxLevel = 0;
    
    char convLimit[256];
    char maxNoDatapoints[256];

    /// Unidim Anova
    char * SplitType = "0";
     
    char maxiter[256];
    char noRestarts[256];
    bool isProbabilistic=false;

    // Put default values in parameters
    trainfilename[0]=0;
    prunefilename[0]=0;
    testfilename[0]=0;
    sprintf(modelfile,"modelfile");
    sprintf(rezfile,"rezfile");
    sprintf(maxiter,"30");
    sprintf(convLimit,"1e-6");
    sprintf(noRestarts,"3");
    sprintf(maxNoDatapoints,"12");

    char c;

    while( (c = getopt ( argc, argv,"t:p:T:m:G:s:PS:r:l:ALQi:c:R:d:h"))!=-1)
        switch(c)
        {
        case 't':
            strcpy(trainfilename,optarg);
            break;

        case 'p':
            strcpy(prunefilename,optarg);
            break;

        case 'T':
            strcpy(testfilename,optarg);
            break;

        case 'm':
            strcpy(modelfile,optarg);
            break;

        case 'G':
            griddim = atoi(optarg);
            break;

        case 's':
            strcpy(rezfile,optarg);
            break;

        case 'P':
            isProbabilistic=true;
            break;

        case 'S':
            c_split = atoi(optarg);
            break;

        case 'r':
            r_dim = atoi(optarg);
            break;

        case 'l':
            MaxLevel = atoi(optarg);
            break;

        case 'A':
            SplitType = "0";
            break;

        case 'L':
            SplitType = "1";
            break;

        case 'Q':
            SplitType = "2";
            break;

        case 'i':
            strcpy(maxiter,optarg);
            break;

        case 'c':
            strcpy(convLimit,optarg);
            break;

        case 'R':
            strcpy(noRestarts,optarg);
            break;

        case 'd':
            strcpy(maxNoDatapoints,optarg);
            break;


        case 'h':
        default:
            cout << "regression: learns a regression tree" << endl;
            cout << "Options:" << endl;
            cout << "\t-h : this help message" << endl;
            cout << "\t-t filename : specify the training set" << endl;
            cout << "\t-p filename : specify the prunning set (absent means no prunning)" << endl;
            cout << "\t-T filename : specify the testing set (absent means no testing)" << endl;
            cout << "\t-m filename : where to put the resulting tree (default=modelfile)" << endl;
            cout << "\t-G number : number of isolines in the grid (no effect if testing)" << endl;
            cout << "\t-s filename : where to put the result of testing" << endl;
            cout << "\t-P : probabilistic regression tree" << endl;
            cout << "\t-S number : numer of continuous split variables" << endl;
            cout << "\t-r number : number of regressor variables" << endl;
            cout << "\t-l number : the biggest nodeID to be used in testing" << endl;
            cout << "\t-A : use unidimentional ANOVA splits" << endl;
            cout << "\t-L : use multidimentional LDA splits" << endl;
            cout << "\t-Q : use multidimentional QDA splits" << endl;
            cout << "\t-i number : max number of iterations in EM algorithm" << endl;
            cout << "\t-c number : convergence tolerance" << endl;
            cout << "\t-R number : number of random restarts for EM" << endl;
            cout << "\t-d number : minimum datapoints in leaf" << endl;

        }

    // Testing the parameters
    if (trainfilename[0]==0)
    {
        cerr << "No training data specified" << endl;
        exit(1);
    }

    bool prune = (prunefilename[0]!=0);
    bool test = (testfilename[0]!=0 || griddim!=0);


    DataProducer*   prod;
    RPMSConsumer*   cons;
    SyncObjList*    gear;
    Machine* rTree;

    try
    {
        srand( time(NULL) );

        gsl_set_error_handler_off(); // make sure that gsl errors do not stop the program

        StreamDCTrainingData* traindata = CreateStreamDCTrainingDataFromFile(trainfilename);
        StreamDCTrainingData* pruningdata = 0;

        if (prune)
            pruningdata = CreateStreamDCTrainingDataFromFile(prunefilename);

        if (!isProbabilistic)
        {
            rTree = new
                    BinaryRegressionTree< MultiDimNormal, LinearRegressor,
                    BinaryObliqueSplitter > ( traindata->GetDDomainSizes(), c_split, r_dim );
        }
        else
        {
            rTree = new
                    BinaryRegressionTree< MultiDimNormal, LinearRegressor,
                    BinaryObliqueProbabilisticSplitter > ( traindata->GetDDomainSizes(), c_split, r_dim );
        }
        /*    BinaryRegressionTree< SkinyMultiDimNormal, LinearRegressor,
        BinaryObliqueSplitter > rTree( traindata->GetDDomainSizes(), c_split, r_dim );
        */

        rTree->SetOption("EMMaxIterations",maxiter);
        rTree->SetOption("EMRestarts",noRestarts);
        rTree->SetOption("ConvergenceLimit",convLimit);
        rTree->SetOption("MaxNoDatapoints",maxNoDatapoints);
        rTree->SetOption("SplitType",SplitType);

        rTree->SetTrainingData(traindata);
        rTree->ScaleData(Scale::Interval);

        rTree->Identify();

        if (prune)
        {
            rTree->SaveToFile("intermfile");
            rTree->SetPruningData(pruningdata);
            rTree->Prune();
        }

        rTree->SaveToFile(modelfile);

        if (test)
        {
            // test now on the same data and put result in testfile
            gear=new SyncObjList();
            if (griddim==0)
            {
                prod = new FileDataProducer(rTree->InDim()+rTree->OutDim(),testfilename);
            }
            else
            {
                prod = new GridInputProducer(griddim,rTree->InDim(), rTree->GetScaleFactors());
            }

            cons=new RPMSConsumer(rTree->InDim(),rezfile,rTree->GetOutput(),
                                  prod->GetOutput());

            rTree->SetInput(prod->GetOutput());
            gear->AddObject(prod);
            gear->AddObject(rTree);
            gear->AddObject(cons);

            char filename[256];

            ofstream out("animation.gpl");

            if (MaxLevel>0)
            {
                for (int maxlevel=0; maxlevel <= MaxLevel; maxlevel++)
                {
                    cout << "Generating resutlt for level: " << maxlevel;

                    sprintf(filename,rezfile,maxlevel);
                    cons->SetFileName(filename);

                    out << "splot \"" << filename << "\"; pause -1" << endl;

                    sprintf(filename,"%d",1<<maxlevel);
                    rTree->SetOption("InferMaxNodeId",filename);
                    gear->Run();
                    cout << "\tMSE=" << cons->GetMSE() << endl;

                }
            }
            else
            {
                cout << "Testing:" << endl;
                rTree->SetOption("InferMaxNodeId","200000000");
                cons->SetFileName(rezfile);
                gear->Run();

                cout << "The error on test data is: " << cons->GetMSE() << endl;
            }
        }
    }
    catch(ErrMsg msg)
    {
        msg.PrintMessage(cout);
    }
    catch(...)
    {
        cerr << "A weird exception cought" << endl;
    }
    return 0;

}

---