001 /**
002 *
003 */
004 package org.wdssii.decisiontree;
005
006 import java.beans.XMLEncoder;
007 import java.io.BufferedReader;
008 import java.io.File;
009 import java.io.FileOutputStream;
010 import java.io.FileReader;
011 import java.io.FileWriter;
012 import java.io.PrintWriter;
013 import java.util.ArrayList;
014 import java.util.Collections;
015 import java.util.List;
016
017 /**
018 * Command-line program that invokes Quinlan's C4.5 algorithm
019 * @author lakshman
020 *
021 */
022 public class Train {
023 /**
024 * Trains and prints out a decision tree
025 */
026 public static void main(String[] args) throws Exception {
027 if (args.length < 2) {
028 System.err
029 .println("Usage: java org.jscience.statistics.decisiontree.Train trainingfile.csv outdir [pruningFraction=0.1] [shuffle=0/1] [separator]\n");
030 System.err
031 .println("The training file should have each example on a line, with all the attributes being numeric and separated by commas. The final attribute should be the true classification\n");
032 System.err
033 .println("Note that if shuffle is turned on, line numbers reported by this program in error messages will be incorrect.");
034 return;
035 }
036 String infile = args[0];
037 String outdir = args[1];
038 String pruningFraction = (args.length > 2) ? args[2] : "0.1";
039 boolean shuffle = (args.length > 3)? (args[3].equals("1")) : false;
040 String separator = (args.length > 4) ? args[4] : ",";
041 BufferedReader reader = null;
042 try {
043 new File(outdir).mkdirs();
044 // read the file
045 reader = new BufferedReader(new FileReader(infile));
046 String line;
047 List<String[]> trainingData = new ArrayList<String[]>();
048 while ((line = reader.readLine()) != null) {
049 String[] splitValues = line.split(separator);
050 trainingData.add(splitValues);
051 }
052
053 if ( shuffle ){
054 Collections.shuffle(trainingData);
055 }
056
057 // get the data and categories
058 int numTraining = trainingData.size();
059 int numAttr = trainingData.get(0).length - 1; // last one is
060 // category
061 float[][] data = new float[numTraining][numAttr];
062 int[] categories = new int[numTraining];
063 for (int i = 0; i < numTraining; ++i) {
064 String[] indata = trainingData.get(i);
065 if (indata.length != (numAttr + 1)) {
066 throw new IllegalArgumentException("Row number " + (i + 1)
067 + " of file has " + indata.length
068 + " attributes. Expected " + (numAttr + 1));
069 }
070 for (int j = 0; j < numAttr; ++j) {
071 try {
072 data[i][j] = Float.parseFloat(indata[j]);
073 } catch (NumberFormatException e) {
074 System.err.println(e.getMessage() + " in attribute no."
075 + (j + 1) + " of line " + (i + 1)
076 + " Assuming zero.");
077 data[i][j] = 0;
078 }
079 }
080 categories[i] = Integer.parseInt(indata[numAttr]);
081 }
082
083
084
085 // Create classifier and learn decision tree
086 QuinlanC45AxialDecisionTreeCreator classifier = new QuinlanC45AxialDecisionTreeCreator(
087 Float.parseFloat(pruningFraction));
088 DecisionTree tree = classifier.learn(data, categories);
089
090 // Java program
091 String program = tree.toJava();
092 System.out.println(program);
093 PrintWriter javaWriter = new PrintWriter(new FileWriter(outdir
094 + "/DecisionTree.java"));
095 javaWriter.println(program);
096 javaWriter.close();
097
098 // skill
099 int[] algResults = new int[categories.length];
100 for (int i = 0; i < algResults.length; ++i) {
101 algResults[i] = tree.classify(data[i]);
102 }
103 MulticategorySkillScore tss = new MulticategorySkillScore(
104 classifier.getNumCategories());
105 tss.update(categories, algResults);
106 System.out.println(tss);
107 PrintWriter skillWriter = new PrintWriter(new FileWriter(outdir
108 + "/skill.txt"));
109 skillWriter.println(tss);
110 skillWriter.close();
111
112 // the tree itself
113 XMLEncoder encoder = new XMLEncoder(new FileOutputStream(outdir
114 + "/decisiontree.xml"));
115 encoder.writeObject(tree);
116 encoder.close();
117 System.out.println("decisiontree.xml written out to " + outdir);
118 } finally {
119 if (reader != null) {
120 reader.close();
121 }
122 }
123 }
124 }