binaryprobabilisticdecisiontreenode.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++ -*-

#ifndef _CLUS_BINARYPROBABILISTICDECISIONTREENODE_H_
#define _CLUS_BINARYPROBABILISTICDECISIONTREENODE_H_

#include "vec.h"

#ifdef DEBUG_PRINT
#include <iostream>
using namespace std;
#endif

using namespace TNT;

namespace CLUS
{

template< class T_Splitter >
class BinaryProbabilisticDecisionTreeNode
{
protected:

    /// unique identifier of the cluster for a regression tree
    int nodeId;
    
    /// the state of the node. At creation em. At load stable
    enum state { stable, split, bootstrap } State;
    
    /// the children of this node
    BinaryProbabilisticDecisionTreeNode< T_Splitter > * Children[2];    

    /// the probability to return first class, for the second probability is 1-probFirstClass
    double probFirstClass;

    /// Splitter for split criterion
    T_Splitter Splitter;

    /// pruning statistics. Their ratio is the error
    double pruningError, pruningTotalMass, pruningTotalMassLeft; 

public:
    BinaryProbabilisticDecisionTreeNode(int NodeId, const Vector<int> & DDomainSize,
                                        int CsplitDim):
            nodeId(NodeId), State(split), Splitter(DDomainSize,CsplitDim)
    {
        probFirstClass=1.0; // predict first class by default
    }

    ~BinaryProbabilisticDecisionTreeNode(void)
    {
        if (Children[0]!=0)
            delete Children[0];
        Children[0]=0;


        if (Children[1]!=0)
            delete Children[1];
        Children[1]=0;
    }

    void StartLearningEpoch(void)
    {
        switch (State)
        {
        case stable:
            if (Children[0]!=0)
            {
                Children[0]->StartLearningEpoch();
                Children[1]->StartLearningEpoch();
            }
            break;
        case split:
            Splitter.InitializeSplitStatistics();
            break;
#if 0

        case bootstrap:
            Splitter.InitializeBootstrapSplitStatistics();
            break;
#endif

        }
    }

    void LearnSample(const int* Dvars, const double* Cvars,
                     int classlabel, double probability, double threshold)
    {
        switch (State)
        {
        case stable:
            {
                // if leaf do nothing
                if (Children[0]==0 || Children[1]==0)
                    return;

                // propagate the learning
                double probabilityLeft = probability*Splitter.ProbabilityLeft(Dvars,Cvars);
                double probabilityRight = probability-probabilityLeft;
                if (probabilityLeft>=threshold)
                {
                    Children[0]->LearnSample(Dvars,Cvars,classlabel,probabilityLeft,threshold);
                }

                if (probabilityRight>=threshold)
                {
                    Children[1]->LearnSample(Dvars,Cvars,classlabel,probabilityRight,threshold);
                }
            }
            break;
        case split:
            Splitter.UpdateSplitStatistics(Dvars, Cvars, classlabel, probability);
            break;
        }
    }

    bool StopLearningEpoch(double minMass)
    {
        switch (State)
        {
        case stable:
            if (Children[0]!=0)
                return Children[0]->StopLearningEpoch(minMass)
                       | Children[1]->StopLearningEpoch(minMass); // || cannot be used since is shortcuted
            else
                return false;

        case split:
            State=stable;

            probFirstClass=Splitter.ComputeProbabilityFirstClass();

            if (!Splitter.ComputeSplitVariable())
            {
                // have to make the node a leaf
                goto make_node_leaf;
            }

#ifdef DEBUG_PRINT
            cout << "NodeId: " << nodeId << " probFirstClass=" << probFirstClass << endl;
#endif

            if (Splitter.MoreSplits(minMass, nodeId))
            {
                Children[0] = new BinaryProbabilisticDecisionTreeNode< T_Splitter >
                              ( nodeId*2, Splitter.GetDDomainSize(), Splitter.GetCSplitDim() );

                Children[1] = new BinaryProbabilisticDecisionTreeNode< T_Splitter >
                              ( nodeId*2+1, Splitter.GetDDomainSize(), Splitter.GetCSplitDim() );
            }
            else
                goto make_node_leaf;

            return true;
            break;

        default:
            return false;
        }

make_node_leaf:
#ifdef DEBUG_PRINT

        cout << "Making the node " << nodeId << " a leaf" << endl;
#endif

        Children[0]=Children[1]=0;
        return false;
    }

    /// Use Baesian decision mode
    double ProbabilityFirstClass(const int* Dvars, const double* Cvars,
                                 double probability, double threshold)
    {
        if (Children[0]==0)
        {
            // leaf node
            return probability*probFirstClass;
        }
        else
        {
            double rez=0.0;
            double probabilityLeft = probability*Splitter.ProbabilityLeft(Dvars,Cvars);
            double probabilityRight = probability-probabilityLeft;

            if (probabilityLeft>=threshold)
            {
                rez+=Children[0]->ProbabilityFirstClass(Dvars,Cvars,probabilityLeft,threshold);
            }

            if (probabilityRight>=threshold)
            {
                rez+=Children[1]->ProbabilityFirstClass(Dvars,Cvars,probabilityRight,threshold);
            }

            return rez;
        }
    }

    void InitializePruningStatistics(void)
    {
        pruningError=0.0;
        pruningTotalMass=0.0;
        pruningTotalMassLeft=0.0;
        if (Children[0]!=0 && Children[1]!=0)
        {
            Children[0]->InitializePruningStatistics();
            Children[1]->InitializePruningStatistics();
        }
    }

    void UpdatePruningStatistics(const int* Dvars, const double* Cvars,  int classlabel /* true output */,
                                 double probability, double threshold)
    {

#if 1
        int predClass = probFirstClass>.5 ? 0 : 1;

        if (classlabel!=predClass)
            pruningError+=probability;
#else

        if (classlabel==0)
        {
            // true result is first class label
            pruningError+=probability*pow2(1.0-probFirstClass);
        }
        else
        {
            pruningError+=probability*pow2(probFirstClass);
        }
#endif

        pruningTotalMass+=probability;

        if (Children[0]==0 || Children[1]==0)
            return; // stop the process

        double probabilityLeft = probability*Splitter.ProbabilityLeft(Dvars,Cvars);
        pruningTotalMassLeft+=probabilityLeft;
        double probabilityRight = probability-probabilityLeft;

        if (probabilityLeft>=threshold*probability)
        {
            Children[0]->UpdatePruningStatistics(Dvars,Cvars,classlabel,probabilityLeft,threshold);
        }

        if (probabilityRight>=threshold*probability)
        {
            Children[1]->UpdatePruningStatistics(Dvars,Cvars,classlabel,probabilityRight,threshold);
        }
    }

    void FinalizePruningStatistics (void)
    {
        // nothing to do
    }

    /// Returns the optimal cost for this subtree and cuts the subtree to optimal size
    double PruneSubtree(void)
    {
        //      double error=pruningTotalMass>0.0 ?
        //pruningError/pruningTotalMass : 0.0;  // normalized error

        double error=pruningError;
#ifdef DEBUG_PRINT

        cout << "Pruneerror of " <<nodeId << " is: " << error
        << " " << error/pruningTotalMass << endl;
#endif

        if (Children[0] == 0 && Children[1] == 0)
        {
            // node is a leaf
            return error;
        }
        else
        {
            //double errorChildren = pruningTotalMassLeft/pruningTotalMass*Children[0]->PruneSubtree()+
            //  (pruningTotalMass-pruningTotalMassLeft)/pruningTotalMass*Children[1]->PruneSubtree();

            double errorChildren = Children[0]->PruneSubtree()+
                                   Children[1]->PruneSubtree();
#ifdef DEBUG_PRINT

            cout << "Childrenerror of " << nodeId << " is: " << errorChildren
            << " " << errorChildren/pruningTotalMass << endl;
#endif

            if (error<=errorChildren)
            {
#ifdef DEBUG_PRINT
                cout << "Prunning at node " << nodeId << endl;
#endif
                // prune the tree here
                delete Children[0];
                Children[0]=0;
                delete Children[1];
                Children[1]=0;

                return error;
            }
            else
            {
                // tree is good as it is
                return errorChildren;
            }
        }
    }

    void SaveToStream(ostream& out)
    {
        out << nodeId << " ";
        //if (Children[0] != 0 && Children[1] != 0)
        Splitter.SaveToStream(out);
        out << " " << probFirstClass << endl;

        if (Children[0] != 0 && Children[1] != 0)
        {
            Children[0]->SaveToStream(out);
            Children[1]->SaveToStream(out);
        }
    }

};
}

#endif // _CLUS_BINARYPROBABILISTICDECISIONTREENODE_H_

---