ai.BalancedRandomTree.java Source code

Java tutorial

Introduction

Here is the source code for ai.BalancedRandomTree.java

Source

package ai;

/**
 *
 * License: GPL
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License 2
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
 *
 * Authors: Ignacio Arganda-Carreras (iarganda@mit.edu), 
 *          Albert Cardona (acardona@ini.phys.ethz.ch)
 */

import ij.IJ;

import java.io.Serializable;
import java.util.ArrayList;
import java.util.LinkedList;

import weka.core.Instance;
import weka.core.Instances;

/**
 * This class implements a random tree based on the split
 * function specified by the template in Splitter
 * 
 */
public class BalancedRandomTree implements Serializable {
    /** Generated serial version UID */
    private static final long serialVersionUID = 41518309467L;
    /** root node */
    private final BaseNode rootNode;

    /**
     * Build random tree for a balanced random forest  
     * 
     * @param data original data
     * @param bagIndices indices of the data samples to use
     * @param splitter split function generator
     */
    public BalancedRandomTree(final Instances data, final ArrayList<Integer> bagIndices, final Splitter splitter) {
        this.rootNode = createNode(data, bagIndices, splitter);
    }

    /**
     * Build the random tree based on the data specified 
     * in the constructor 
     */
    private final BaseNode createNode(final Instances data, final ArrayList<Integer> bagIndices,
            final Splitter splitter) {
        final long start = System.currentTimeMillis();
        try {
            return createTree(data, bagIndices, 0, splitter);
        } finally {
            final long end = System.currentTimeMillis();
            IJ.log("Creating tree took: " + (end - start) + "ms");
        }
    }

    /**
     * Evaluate sample
     * 
     * @param instance sample to evaluate
     * @return array of class probabilities
     */
    public double[] evaluate(Instance instance) {
        if (null == rootNode)
            return null;
        return rootNode.eval(instance);
    }

    /**
     * Basic node of the tree
     *
     */
    abstract class BaseNode implements Serializable {

        /** serial version ID */
        private static final long serialVersionUID = 46734234231L;

        /**
         * Evaluate an instance
         * @param instance input sample
         * @return class probabilities
         */
        public abstract double[] eval(Instance instance);

        /**
         * Get the node depth
         * 
         * @return tree depth at that node
         */
        public int getDepth() {
            return 0;
        }
    } // end class BaseNode

    /**
     * Leaf node in the tree 
     *
     */
    class LeafNode extends BaseNode implements Serializable {
        /** serial version ID */
        private static final long serialVersionUID = 2019873470157L;
        /**class probabilites */
        double[] probability;

        @Override
        public double[] eval(Instance instance) {
            return probability;
        }

        /**
         * Create a leaf node
         * 
         * @param probability class probabilities
         */
        public LeafNode(double[] probability) {
            this.probability = probability;
        }

        /**
         * Create leaf node based on the current split data
         *  
         * @param data pointer to original data
         * @param indices indices at this node
         */
        public LeafNode(final Instances data, ArrayList<Integer> indices) {
            this.probability = new double[data.numClasses()];
            for (final Integer it : indices) {
                this.probability[(int) data.get(it.intValue()).classValue()]++;
            }
            // Divide by the number of elements
            for (int i = 0; i < data.numClasses(); i++)
                this.probability[i] /= (double) indices.size();
        }

    } //end class LeafNode

    /**
     * Interior node of the tree
     *
     */
    class InteriorNode extends BaseNode implements Serializable {
        /** serial version ID */
        private static final long serialVersionUID = 9972970234021L;
        /** left son */
        BaseNode left;
        /** right son */
        BaseNode right;
        /** node depth */
        final int depth;
        /** split function that divides the samples into left and right sons */
        final SplitFunction splitFn;

        /**
         * Constructs an interior node of the random tree
         * 
         * @param depth tree depth at this node
         * @param splitFn split function
         */
        private InteriorNode(int depth, SplitFunction splitFn) {
            this.depth = depth;
            this.splitFn = splitFn;
        }

        /**
         * Construct interior node of the tree
         * 
         * @param data pointer to the original set of samples
         * @param indices indices of the samples at this node
         * @param depth current tree depth
         */
        /*      public InteriorNode(
        final Instances data,
        final ArrayList<Integer> indices,
        final int depth,
        final Splitter splitFnProducer)
        {
           this.splitFn = splitFnProducer.getSplitFunction(data, indices);
            
            
           this.depth = depth;
            
           // left and right new arrays
           final ArrayList<Integer> leftArray = new ArrayList<Integer>();
           final ArrayList<Integer> rightArray = new ArrayList<Integer>();
            
           // split data
           int totalLeft = 0;
           int totalRight = 0;
           for(final Integer it : indices)
           {
        if( splitFn.evaluate( data.get(it.intValue()) ) )
        {
           leftArray.add(it);
           totalLeft ++;               
        }
        else
        {
           rightArray.add(it);
           totalRight ++;
        }
           }
           //System.out.println("total left = " + totalLeft + ", total rigth = " + totalRight + ", depth = " + depth);               
           //indices.clear();
           if( totalLeft == 0 )
           {
        left = new LeafNode(data, rightArray);
           }
           else if ( totalRight == 0 )
           {
        left = new LeafNode(data, leftArray);
           }
           else
           {
        left = new InteriorNode(data, leftArray, depth+1, splitFnProducer);
        right = new InteriorNode(data, rightArray, depth+1, splitFnProducer);
           }            
        }*/

        /**
         * Evaluate sample at this node
         */
        public double[] eval(Instance instance) {
            if (null != right) {
                if (this.splitFn.evaluate(instance)) {
                    return left.eval(instance);
                } else
                    return right.eval(instance);
            } else // leaves are always left nodes 
                return left.eval(instance);
        }

        /**
         * Get node depth
         */
        public int getDepth() {
            return this.depth;
        }
    }

    /**
     * Create random tree (non-recursively)
     * 
     * @param data original data
     * @param indices indices of the samples to use
     * @param depth starting depth
     * @param splitFnProducer split function producer
     * @return root node 
     */
    private InteriorNode createTree(final Instances data, final ArrayList<Integer> indices, final int depth,
            final Splitter splitFnProducer) {
        int maxDepth = depth;
        // Create root node
        InteriorNode root = new InteriorNode(depth, splitFnProducer.getSplitFunction(data, indices));

        // Create list of nodes to process and add the root to it
        final LinkedList<InteriorNode> remainingNodes = new LinkedList<InteriorNode>();
        remainingNodes.add(root);

        // Create list of indices to process (it must match all the time with the node list)
        final LinkedList<ArrayList<Integer>> remainingIndices = new LinkedList<ArrayList<Integer>>();
        remainingIndices.add(indices);

        // While there is still nodes to process
        while (!remainingNodes.isEmpty()) {
            final InteriorNode currentNode = remainingNodes.removeLast();
            final ArrayList<Integer> currentIndices = remainingIndices.removeLast();
            // new arrays of indices for the left and right sons
            final ArrayList<Integer> leftArray = new ArrayList<Integer>();
            final ArrayList<Integer> rightArray = new ArrayList<Integer>();

            // split data
            for (final Integer it : currentIndices) {
                if (currentNode.splitFn.evaluate(data.get(it.intValue()))) {
                    leftArray.add(it);
                } else {
                    rightArray.add(it);
                }
            }
            //System.out.println("total left = " + leftArray.size() + ", total right = " + rightArray.size() + ", depth = " + currentNode.depth);               
            // Update maximum depth (for the record)
            if (currentNode.depth > maxDepth)
                maxDepth = currentNode.depth;

            if (leftArray.isEmpty()) {
                currentNode.left = new LeafNode(data, rightArray);
                //System.out.println("Created leaf with feature " + currentNode.splitFn.index);
            } else if (rightArray.isEmpty()) {
                currentNode.left = new LeafNode(data, leftArray);
                //System.out.println("Created leaf with feature " + currentNode.splitFn.index);
            } else {
                currentNode.left = new InteriorNode(currentNode.depth + 1,
                        splitFnProducer.getSplitFunction(data, leftArray));
                remainingNodes.add((InteriorNode) currentNode.left);
                remainingIndices.add(leftArray);

                currentNode.right = new InteriorNode(currentNode.depth + 1,
                        splitFnProducer.getSplitFunction(data, rightArray));
                remainingNodes.add((InteriorNode) currentNode.right);
                remainingIndices.add(rightArray);
            }
        }

        System.out.println("Max depth = " + maxDepth);
        return root;
    }

}