binarydecisiontreenode.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_BINARYDECISIONTREENODE_H_
#define _CLUS_BINARYDECISIONTREENODE_H_

#include "vec.h"
#include "general.h"

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

using namespace TNT;

namespace CLUS
{

/// Implements a node of the binary decision tree
template< class T_Splitter >
class BinaryDecisionTreeNode
{
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} State;

    /// the predicted class label
    int classLabel;

    /// the children of this node
    BinaryDecisionTreeNode< T_Splitter > * Children[2];

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

    /// splitter for split criterion
    T_Splitter Splitter;

    /// pruning error statistic
    int pruningError;

    /// pruning total mass statistic
    int pruningTotalMass;

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

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


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

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

    /// Learn a data sample
    ///
    /// @param Dvars         discrete variables
    /// @param Cvars         continuous variables
    /// @param classlabel    classification label
    void LearnSample(const int* Dvars, const double* Cvars, int classlabel)
    {
        switch (State)
        {
        case stable:
            // if leaf do nothing
            if (Children[0]==0 || Children[1]==0)
                return;

            if (Splitter.ChooseBranch(Dvars,Cvars)==0)
                Children[0]->LearnSample(Dvars,Cvars,classlabel);
            else
                Children[1]->LearnSample(Dvars,Cvars,classlabel);
            break;

        case split:
            Splitter.UpdateSplitStatistics(Dvars, Cvars, classlabel);
            break;
        }
    }

    /// Stop the learning process
    ///
    /// @return              true if stopped successfully
    ///                      false if still processing data
    bool StopLearningEpoch(int 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;

            classLabel=Splitter.ComputeClassLabel();

            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 BinaryDecisionTreeNode< T_Splitter >
                              ( nodeId*2, Splitter.GetDDomainSize(), Splitter.GetCSplitDim() );

                Children[1] = new BinaryDecisionTreeNode< 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;
    }

    /// Do the inference
    ///
    /// @param Dvars         discrete variables
    /// @param Cvars         continuous variables
    double Infer(const int* Dvars, const double* Cvars)
    {
        // cout << "I am node: " << nodeId << " and my classlabel is " << classLabel << endl;
        if (Children[0]==0 || Children[1]==0)
        {
            // leaf node
            return classLabel;
        }
        else
        {
            return Children[Splitter.ChooseBranch(Dvars,Cvars)]->Infer(Dvars,Cvars);
        }
    }

    /// Initialize stats about pruning
    void InitializePruningStatistics(void)
    {
        pruningError=0;
        pruningTotalMass=0;

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

    /// Update pruning stats with new data
    ///
    /// @param Dvars         discrete variables
    /// @param Cvars         continuous variables
    /// @param classlabel    classification label
    void UpdatePruningStatistics(const int* Dvars, const double* Cvars,
                                 int classlabel /* true output */)
    {

        if (classlabel!=classLabel)
            pruningError+=1;

        pruningTotalMass+=1;

        if (Children[0]==0 || Children[1]==0)
            return; // stop the process
        else
            Children[Splitter.ChooseBranch(Dvars,Cvars)]->UpdatePruningStatistics(Dvars,Cvars,classlabel);

    }

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

    /// Return the optimal cost for this subtree and cuts the subtree to optimal size
    double PruneSubtree(void)
    {
#ifdef DEBUG_PRINT
        cout << "Pruneerror of " <<nodeId << " is: " << pruningError
        << " " << 1.0*pruningError/pruningTotalMass << endl;
#endif

        if (Children[0] == 0 && Children[1] == 0)
        {
            // node is a leaf
            return pruningError;
        }
        else
        {
            double errorChildren = Children[0]->PruneSubtree()+
                                   Children[1]->PruneSubtree();
#ifdef DEBUG_PRINT

            cout << "Childrenerror of " << nodeId << " is: " << errorChildren
            << " " << 1.0*errorChildren/pruningTotalMass << endl;
#endif

            if (pruningError<=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 pruningError;
            }
            else
            {
                // tree is good as it is
                return errorChildren;
            }
        }
    }

    /// Output the node data to a stream
    ///
    /// @param out           stream for output
    void SaveToStream(ostream& out)
    {
        out << nodeId << " ";
        //if (Children[0] != 0 && Children[1] != 0)
        Splitter.SaveToStream(out);
        out << " " << classLabel << endl;

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

};
}

#endif // _CLUS_BINARYDECISIONTREENODE_H_

---