regressiontree.h

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.

*/

// -*- C++ -*-

#if !defined _CLUS_REGRESSIONTREE_H_
#define _CLUS_REGRESSIONTREE_H_

#include "machine.h"
#include "regressiontreenode.h"
#include "dctraingen.h"
#include <iostream>

// using namespace TNT;

namespace CLUS
{

template< class T_Distribution, class T_Regressor, class T_Splitter >
class BinaryRegressionTree : public Machine
{
protected:
    BinaryRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter>* root;

    /// list of discrete domain sizes
    const Vector<int>& dDomainSize;

    /// num of discrete variables
    int dsplitDim;

    /// num of continuous+split variables
    int csplitDim;

    /// num of regression variables
    int regDim;

    /// num of iterations to get convergence for EM
    int emMaxIterations;

    /// num of restarts of EM to get a good initial starting point
    int emRestarts;

    /// the minimum number of datapoints in a node to split further
    int min_no_datapoints;

    /// type of split to be passed to splitter, splitter dependent
    int splitType;

    /// threshold for considering a branch irrelevant
    double threshold; 

    T_Distribution* rootDistribution;
    int inferMaxNodeId;

    void PrintSizesTree(void)
    {
        int nodes=0;
        int term_nodes=0;
        root->ComputeSizesTree(nodes,term_nodes);
        cout << "Nuber of nodes=" << nodes << "\tNumber of terminal nodes=" << term_nodes << endl;
    }

public:
    BinaryRegressionTree(const Vector<int>& DDomainSize, int CsplitDim, int RegDim):
            Machine(CsplitDim+RegDim,1),dDomainSize(DDomainSize),
            dsplitDim(DDomainSize.dim()),csplitDim(CsplitDim), regDim(RegDim)
    {
        emRestarts = 3;
        emMaxIterations = 30;
        min_no_datapoints = 10;
        splitType = 0;
        rootDistribution = 0;
        inferMaxNodeId = INT_MAX;
        threshold=.01;
    }

    virtual ~BinaryRegressionTree(void)
    {
        if (rootDistribution)
            delete rootDistribution;
    }

    virtual int InDim(void)
    {
        return dsplitDim+csplitDim+regDim;
    }

    virtual string TypeName(void)
    {
        return string("BinaryRegressionTree");
    }

    virtual void Infer(void)
    {
        if (root==0)
            return;
            
        // translate the first dsplitDim inputs into int
        int Dvars[MAX_VARIABLES];
        for(int i=0; i<dsplitDim; i++)
            Dvars[i]=(int)(*input)[i];
            
        // scale the continuous inputs
        double scaledInput[MAX_VARIABLES];
        for (int i=0; i<csplitDim+regDim; i++)
            scaledInput[i]=scale[i].Transform( (*input)[i+dsplitDim] );

        output[0]=scale[csplitDim+regDim].Transform( root->Infer(Dvars,scaledInput,inferMaxNodeId, threshold) );
    }

    virtual void Identify(void)
    {
        const Matrix<double>& ctrainData = training->GetTrainingData();
        const Matrix<int>& dtrainData = dynamic_cast< DCTrainingData* >( training )
                                        -> GetDiscreteTrainingData();

        int M=ctrainData.num_rows();

        // find a linear regressor for the first node
        if (!rootDistribution)
            rootDistribution = new T_Distribution(regDim);

        rootDistribution->RandomDistribution(1);
        double Coef;
        for (int i=0; i<M; i++)
        {
            Coef = rootDistribution->LearnProbability(ctrainData[i]+csplitDim);
            rootDistribution->NormalizeLearnProbability(Coef);
        }
        rootDistribution->UpdateParameters();

        /*
        cout << "Printing root distribution" << endl;
        rootDistribution->SaveToStream(cout);
        cout << endl;
        */

        T_Regressor* regressor = dynamic_cast<T_Regressor*>( rootDistribution->CreateRegressor() );

        if (regressor==NULL)
            regressor=new T_Regressor();

        // create the root and give it the Id 1
        root = new BinaryRegressionTreeNode<T_Distribution, T_Regressor, T_Splitter>
               (1, dDomainSize, csplitDim, regDim, *regressor, rootDistribution);

        rootDistribution = 0; // root will dealocate rootDistribution
        // learn in stages until nobody wants to learn anymore
        do
        {
            root->StartLearningEpoch();
            for(int i=0; i<M; i++)
                root->LearnSample(dtrainData[i],ctrainData[i], 1.0, threshold);
        }
        while (root->StopLearningEpoch(splitType, emRestarts, emMaxIterations,
                                       convergenceLim, min_no_datapoints));
        cout << "End Learning" << endl;
        PrintSizesTree();
    }

    virtual void Prune(void)
    {
        const Matrix<double>& ctrainData = pruning->GetTrainingData();
        const Matrix<int>& dtrainData = dynamic_cast< DCTrainingData* >( pruning )
                                        -> GetDiscreteTrainingData();

        int M=ctrainData.num_rows();

        root->InitializePruningStatistics();
        for(int i=0; i<M; i++)
        {
            // scale the continuous inputs
            double scaledInput[MAX_VARIABLES];
            for (int j=0; j<csplitDim+regDim; j++)
                scaledInput[j]=scale[j].Transform( ctrainData[i][j] );
            // scale the output
            double y=scale[csplitDim+regDim].InverseTransform( ctrainData[i][csplitDim+regDim] );

            root->UpdatePruningStatistics(dtrainData[i], scaledInput, y, 1.0, threshold);
        }
        root->FinalizePruningStatistics();

        // now cut the tree to the right size
        double cost=root->PruneSubtree();
        cout << "RMSN after pruning is:" << cost/M << endl;
        PrintSizesTree();
    }

    virtual int SetOption(char* name, char* val)
    {
        if (strcmp(name,"EMMaxIterations")==0)
            emMaxIterations = atoi(val);
        else
            if (strcmp(name,"EMRestarts")==0)
                emRestarts = atoi(val);
            else
                if (strcmp(name,"InferMaxNodeId")==0)
                    inferMaxNodeId = atoi(val);
                else
                    if (strcmp(name,"MaxNoDatapoints")==0)
                        min_no_datapoints = 2*atoi(val)-2;
                    else
                        if (strcmp(name,"SplitType")==0)
                            splitType = atoi(val);
                        else
                            if (strcmp(name,"Threshold")==0)
                                threshold = atof(val);
                            else
                                return Machine::SetOption(name,val);
        return 1;
    }

    virtual void SaveToStream(ostream& out)
    {
        out << TypeName() << " (  " << dsplitDim << " " << csplitDim << " ";
        out << regDim << " ) { "  << endl;

        out << "[ ";
        for(int i=0; i<dsplitDim; i++)
            out << dDomainSize[i] << " ";
        out << "]" << endl;

        for(int i=0; i<csplitDim+regDim+1; i++)
            scale[i].SaveToStream(out);
        out << endl;

        root->SaveToStream(out);

        out << '}' << endl;
    }
};

}

#endif // _CLUS_REGRESSIONTREE_H_

---