package weka.classifiers;
import java.util.*;
import java.io.*;
import weka.core.*;
import weka.estimators.*;
import java.util.zip.GZIPInputStream;
import java.util.zip.GZIPOutputStream;
/**
* Class for evaluating machine learning models. <p>
* Example usage as the main of a classifier (called FunkyClassifier):
* <code> <pre>
* public static void main(String [] args) {
* try {
* Classifier scheme = new FunkyClassifier();
* System.out.println(Evaluation.evaluateModel(scheme, args));
* } catch (Exception e) {
* System.err.println(e.getMessage());
* }
* }
* </pre> </code>
* <p>
* Example usage from within an application:
* <code> <pre>
* Instances trainInstances = ... instances got from somewhere
* Instances testInstances = ... instances got from somewhere
* Classifier scheme = ... scheme got from somewhere
*
* Evaluation evaluation = new Evaluation(trainInstances);
* evaluation.evaluateModel(scheme, testInstances);
* System.out.println(evaluation.toSummaryString());
* </pre> </code>
*/
public class Evaluation implements Summarizable {
private Instances template;
/** The number of classes. */
private int m_NumClasses;
/** The number of folds for a cross-validation. */
private int m_NumFolds;
/** The weight of all incorrectly classified instances. */
private double m_Incorrect;
/** The weight of all correctly classified instances. */
private double m_Correct;
/** The weight of all unclassified instances. */
private double m_Unclassified;
/*** The weight of all instances that had no class assigned to them. */
private double m_MissingClass;
/** The weight of all instances that had a class assigned to them. */
private double m_WithClass;
/** Array for storing the confusion matrix. */
private double [][] m_ConfusionMatrix;
/** The names of the classes. */
private String [] m_ClassNames;
/** Is the class nominal or numeric? */
private boolean m_ClassIsNominal;
/** The prior probabilities of the classes */
private double [] m_ClassPriors;
/** The sum of counts for priors */
private double m_ClassPriorsSum;
/** The cost matrix (if given). */
private CostMatrix m_CostMatrix;
/** The total cost of predictions (includes instance weights) */
private double m_TotalCost;
/** Sum of errors. */
private double m_SumErr;
/** Sum of absolute errors. */
private double m_SumAbsErr;
/** Sum of squared errors. */
private double m_SumSqrErr;
/** Sum of class values. */
private double m_SumClass;
/** Sum of squared class values. */
private double m_SumSqrClass;
/*** Sum of predicted values. */
private double m_SumPredicted;
/** Sum of squared predicted values. */
private double m_SumSqrPredicted;
/** Sum of predicted * class values. */
private double m_SumClassPredicted;
/** Sum of absolute errors of the prior */
private double m_SumPriorAbsErr;
/** Sum of absolute errors of the prior */
private double m_SumPriorSqrErr;
/** Total Kononenko & Bratko Information */
private double m_SumKBInfo;
/*** Resolution of the margin histogram */
private static int k_MarginResolution = 500;
/** Cumulative margin distribution */
private double m_MarginCounts [];
/** Number of non-missing class training instances seen */
private int m_NumTrainClassVals;
/** Array containing all numeric training class values seen */
private double [] m_TrainClassVals;
/** Array containing all numeric training class weights */
private double [] m_TrainClassWeights;
/** Numeric class error estimator for prior */
private Estimator m_PriorErrorEstimator;
/** Numeric class error estimator for scheme */
private Estimator m_ErrorEstimator;
/**
* The minimum probablility accepted from an estimator to avoid taking log(0) in Sf calculations.
*/
private static final double MIN_SF_PROB = Double.MIN_VALUE;
/** Total entropy of prior predictions */
private double m_SumPriorEntropy;
/** Total entropy of scheme predictions */
private double m_SumSchemeEntropy;
/**
* Initializes all the counters for the evaluation.
* @param data set of training instances, to get some header
* information and prior class distribution information
* @exception Exception if the class is not defined
*/
public Evaluation(Instances data) throws Exception {
this(data, null);
}
/**
* Initializes all the counters for the evaluation and also takes a
* cost matrix as parameter.
*
* @param data set of instances, to get some header information
* @param costMatrix the cost matrix---if null, default costs will be used
* @exception Exception if cost matrix is not compatible with
* data, the class is not defined or the class is numeric
*/
public Evaluation(Instances data,CostMatrix costMatrix)throws Exception{
template=data;
m_NumClasses = data.numClasses();
m_NumFolds = 1;
m_ClassIsNominal = data.classAttribute().isNominal();
if (m_ClassIsNominal) {
m_ConfusionMatrix = new double [m_NumClasses][m_NumClasses];
m_ClassNames = new String [m_NumClasses];
for(int i = 0; i < m_NumClasses; i++) {
m_ClassNames[i] = data.classAttribute().value(i);
}
}
m_CostMatrix = costMatrix;
if (m_CostMatrix != null) {
if (!m_ClassIsNominal) {
throw new Exception("Class has to be nominal if cost matrix " +
"given!");
}
if (m_CostMatrix.size() != m_NumClasses) {
throw new Exception("Cost matrix not compatible with data!");
}
}
m_ClassPriors = new double [m_NumClasses];
setPriors(data);
m_MarginCounts = new double [k_MarginResolution + 1];
}
/**
* Returns a copy of the confusion matrix.
* @return a copy of the confusion matrix as a two-dimensional array
*/
public double[][] confusionMatrix() {
return Utils.clone(m_ConfusionMatrix);
}
/**
* Performs a (stratified if class is nominal) cross-validation
* for a classifier on a set of instances.
* @param classifier the classifier with any options set.
* @param data the data on which the cross-validation is to be
* performed
* @param numFolds the number of folds for the cross-validation
* @exception Exception if a classifier could not be generated
* successfully or the class is not defined
*/
public void crossValidateModel(Classifier classifier,Instances data,int numFolds)throws Exception{
// Make a copy of the data we can reorder
data = new Instances(data);
if (data.classAttribute().isNominal()) {
data.stratify(numFolds);
}
// Do the folds
for (int i = 0; i < numFolds; i++) {
Instances train = data.trainCV(numFolds, i);
setPriors(train);
classifier.buildClassifier(train);
Instances test = data.testCV(numFolds, i);
evaluateModel(classifier, test);
}
m_NumFolds = numFolds;
}
/**
* Performs a (stratified if class is nominal) cross-validation
* for a classifier on a set of instances.
* @param classifier a string naming the class of the classifier
* @param data the data on which the cross-validation is to be
* performed
* @param numFolds the number of folds for the cross-validation
* @param options the options to the classifier. Any options
* accepted by the classifier will be removed from this array.
* @exception Exception if a classifier could not be generated
* successfully or the class is not defined
*/
public void crossValidateModel(String classifierString,Instances data,int numFolds,String[] options)throws Exception{
crossValidateModel(Classifier.forName(classifierString, options),data,numFolds);
}
/**
* Evaluates a classifier with the options given in an array of strings.
* @param classifierString class of machine learning classifier as a string
* @param options the array of string containing the options
* @exception Exception if model could not be evaluated successfully
* @return a string describing the results
*/
public static String evaluateModel(String classifierString,String [] options)throws Exception{
Classifier classifier;
try {
classifier=(Classifier)Class.forName(classifierString).newInstance();
} catch (Exception e) {
throw new Exception("Can't find class with name "+classifierString+'.');
}
return evaluateModel(classifier, options);
}
/**
* A test method for this class. Just extracts the first command line argument as a classifier class name and calls evaluateModel.
* @param args an array of command line arguments, the first of which must be the class name of a classifier.
*/
public static void main(String [] args) {
try {
if(args.length==0)throw new Exception("The first argument must be the class name"+" of a classifier");
String classifier = args[0];
args[0] = "";
System.out.println(evaluateModel(classifier, args));
} catch (Exception ex) {
ex.printStackTrace();
System.err.println(ex.getMessage());
}
}
/**
* Evaluates a classifier with the options given in an array of
* strings. <p>
*
* Valid options are: <p>
*
* -t name of training file <br>
* Name of the file with the training data. (required) <p>
*
* -T name of test file <br>
* Name of the file with the test data. If missing a cross-validation is performed. <p>
*
* -c class index <br>
* Index of the class attribute (1, 2, ...; default: last). <p>
*
* -x number of folds <br>
* The number of folds for the cross-validation (default: 10). <p>
*
* -s random number seed <br>
* Random number seed for the cross-validation (default: 1). <p>
*
* -m file with cost matrix <br>
* The name of a file containing a cost matrix. <p>
*
* -l name of model input file <br>
* Loads classifier from the given file. <p>
*
* -d name of model output file <br>
* Saves classifier built from the training data into the given file. <p>
*
* -v <br>
* Outputs no statistics for the training data. <p>
*
* -o <br>
* Outputs statistics only, not the classifier. <p>
*
* -i <br>
* Outputs detailed information-retrieval statistics per class. <p>
*
* -k <br>
* Outputs information-theoretic statistics. <p>
*
* -p <br>
* Expects a parameter specifying a range of attributes to list with the predictions. <p>
*
* -P name of result file <br>
* Name of the file where the predictions for test instances to be output. <p>
*
* -r <br>
* Outputs cumulative margin distribution (and nothing else). <p>
*
* -g <br>
* Only for classifiers that implement "Graphable." Outputs
* the graph representation of the classifier (and nothing
* else). <p>
*
* @param classifier machine learning classifier
* @param options the array of string containing the options
* @exception Exception if model could not be evaluated successfully
* @return a string describing the results */
public static String evaluateModel(Classifier classifier,String [] options)throws Exception{
Instances train = null, tempTrain, test = null, template = null;
int seed = 1, folds = 10, classIndex = -1;
String trainFileName, testFileName, sourceClass, classIndexString, objectInputFileName, objectOutputFileName, attributeRangeString, predictionFileName;
boolean IRstatistics = false, noOutput = false, trainStatistics = true, printMargins = false, printComplexityStatistics = false, printGraph = false, classStatistics = false, printSource = false;
StringBuffer text = new StringBuffer();
BufferedReader trainReader = null, testReader = null;
ObjectInputStream objectInputStream = null;
Random random;
CostMatrix costMatrix = null;
Range attributesToOutput = null;
long trainTimeStart = 0, trainTimeElapsed = 0, testTimeStart = 0, testTimeElapsed = 0;
try {
// Get basic options (options the same for all schemes)
classIndexString = Utils.getOption('c', options);
if(classIndexString.length()!=0)classIndex=Integer.parseInt(classIndexString);
trainFileName = Utils.getOption('t', options);
objectInputFileName = Utils.getOption('l', options);
objectOutputFileName = Utils.getOption('d', options);
testFileName = Utils.getOption('T', options);
if (trainFileName.length() == 0) {
if (objectInputFileName.length() == 0) {
throw new Exception("No training file and no object "+
"input file given.");
}
if (testFileName.length() == 0) {
throw new Exception("No training file and no test "+
"file given.");
}
} else if ((objectInputFileName.length() != 0) &&
((!(classifier instanceof UpdateableClassifier)) ||
(testFileName.length() == 0))) {
throw new Exception("Classifier not incremental, or no " +
"test file provided: can't "+
"use both train and model file.");
}
try {
if (trainFileName.length() != 0) {
trainReader = new BufferedReader(new FileReader(trainFileName));
}
if (testFileName.length() != 0) {
testReader = new BufferedReader(new FileReader(testFileName));
}
if (objectInputFileName.length() != 0) {
InputStream is = new FileInputStream(objectInputFileName);
if (objectInputFileName.endsWith(".gz")) {
is = new GZIPInputStream(is);
}
objectInputStream = new ObjectInputStream(is);
}
} catch (Exception e) {
throw new Exception("Can't open file " + e.getMessage() + '.');
}
if(testFileName.length()!=0){
template=test=new Instances(testReader,1);
if(classIndex!=-1)test.setClassIndex(classIndex-1);
else test.setClassIndex(test.numAttributes()-1);
}
if(trainFileName.length()!=0){
if((classifier instanceof UpdateableClassifier)&&(testFileName.length()!=0))train=new Instances(trainReader,1);
else train=new Instances(trainReader);
template=train;
if(classIndex!=-1)train.setClassIndex(classIndex-1);
else train.setClassIndex(train.numAttributes()-1);
if(classIndex>train.numAttributes())throw new Exception("Index of class attribute too large.");
}
if(template==null)throw new Exception("No actual dataset provided to use as template");
seed=Utils.getInt('s',seed,options);
folds=Utils.getInt('x',folds,options);
costMatrix = handleCostOption(Utils.getOption('m', options), template.numClasses());
classStatistics = Utils.getFlag('i', options);
noOutput = Utils.getFlag('o', options);
trainStatistics = !Utils.getFlag('v', options);
printComplexityStatistics = Utils.getFlag('k', options);
printMargins = Utils.getFlag('r', options);
printGraph = Utils.getFlag('g', options);
sourceClass = Utils.getOption('z', options);
printSource = (sourceClass.length() != 0);
attributeRangeString = Utils.getOption('p', options);
if(attributeRangeString.length()!=0)attributesToOutput=new Range(attributeRangeString);
predictionFileName=Utils.getOption('P',options);
// If a model file is given, we can't process scheme-specific options
if(objectInputFileName.length()==0&&classifier instanceof OptionHandler)((OptionHandler)classifier).setOptions(options);
Utils.checkForRemainingOptions(options);
} catch (Exception e) {
throw new Exception("\nWeka exception: " + e.getMessage()
+ makeOptionString(classifier));
}
// Setup up evaluation objects
Evaluation trainingEvaluation = new Evaluation(new Instances(template, 0), costMatrix);
Evaluation testingEvaluation = new Evaluation(new Instances(template, 0), costMatrix);
if(objectInputFileName.length()!=0){
// Load classifier from file
classifier = (Classifier) objectInputStream.readObject();
objectInputStream.close();
}
// Build the classifier if no object file provided
if ((classifier instanceof UpdateableClassifier) &&
(testFileName.length() != 0) &&
(costMatrix == null) &&
(trainFileName.length() != 0)) {
// Build classifier incrementally
trainingEvaluation.setPriors(train);
testingEvaluation.setPriors(train);
trainTimeStart = System.currentTimeMillis();
if (objectInputFileName.length() == 0) {
classifier.buildClassifier(train);
}
while (train.readInstance(trainReader)) {
trainingEvaluation.updatePriors(train.instance(0));
testingEvaluation.updatePriors(train.instance(0));
((UpdateableClassifier)classifier).
updateClassifier(train.instance(0));
train.delete(0);
}
trainTimeElapsed = System.currentTimeMillis() - trainTimeStart;
trainReader.close();
} else if (objectInputFileName.length() == 0) {
// Build classifier in one go
tempTrain = new Instances(train);
trainingEvaluation.setPriors(tempTrain);
testingEvaluation.setPriors(tempTrain);
trainTimeStart = System.currentTimeMillis();
classifier.buildClassifier(tempTrain);
trainTimeElapsed = System.currentTimeMillis() - trainTimeStart;
}
// Save the classifier if an object output file is provided
if (objectOutputFileName.length() != 0) {
OutputStream os = new FileOutputStream(objectOutputFileName);
if (objectOutputFileName.endsWith(".gz")) {
os = new GZIPOutputStream(os);
}
ObjectOutputStream objectOutputStream = new ObjectOutputStream(os);
objectOutputStream.writeObject(classifier);
objectOutputStream.flush();
objectOutputStream.close();
}
// If classifier is drawable output string describing graph
if ((classifier instanceof Drawable)
&& (printGraph)){
return ((Drawable)classifier).graph();
}
// Output the classifier as equivalent source
if ((classifier instanceof Sourcable)
&& (printSource)){
return wekaStaticWrapper((Sourcable) classifier, sourceClass);
}
if(predictionFileName.length()>0){
PrintWriter writer=new PrintWriter(new FileOutputStream(predictionFileName));
writer.print(printClassifications(classifier,new Instances(template,0),testFileName,classIndex,attributesToOutput));
writer.close();
}
// Output model
if (!(noOutput || printMargins)) {
text.append("\n" + classifier.toString() + "\n");
}
if (!printMargins && (costMatrix != null)) {
text.append("\n=== Evaluation Cost Matrix ===\n\n")
.append(costMatrix.toString());
}
// Compute error estimate from training data
if ((trainStatistics) &&
(trainFileName.length() != 0)) {
if ((classifier instanceof UpdateableClassifier) &&
(testFileName.length() != 0) &&
(costMatrix == null)) {
// Classifier was trained incrementally, so we have to
// reopen the training data in order to test on it.
trainReader = new BufferedReader(new FileReader(trainFileName));
// Incremental testing
train = new Instances(trainReader, 1);
if (classIndex != -1) {
train.setClassIndex(classIndex - 1);
} else {
train.setClassIndex(train.numAttributes() - 1);
}
testTimeStart = System.currentTimeMillis();
while (train.readInstance(trainReader)) {
trainingEvaluation.
evaluateModelOnce((Classifier)classifier,
train.instance(0));
train.delete(0);
}
testTimeElapsed = System.currentTimeMillis() - testTimeStart;
trainReader.close();
} else {
testTimeStart = System.currentTimeMillis();
trainingEvaluation.evaluateModel(classifier,
train);
testTimeElapsed = System.currentTimeMillis() - testTimeStart;
}
// Print the results of the training evaluation
if (printMargins) {
return trainingEvaluation.toCumulativeMarginDistributionString();
} else {
text.append("\nTime taken to build model: " +
Utils.doubleToString(trainTimeElapsed / 1000.0,2) +
" seconds");
text.append("\nTime taken to test model on training data: " +
Utils.doubleToString(testTimeElapsed / 1000.0,2) +
" seconds");
text.append(trainingEvaluation.
toSummaryString("\n\n=== Error on training" +
" data ===\n", printComplexityStatistics));
if (template.classAttribute().isNominal()) {
if (classStatistics) {
text.append("\n\n" + trainingEvaluation.toClassDetailsString());
}
text.append("\n\n" + trainingEvaluation.toMatrixString());
}
}
}
// Compute proper error estimates
if (testFileName.length() != 0) {
// Testing is on the supplied test data
while (test.readInstance(testReader)) {
testingEvaluation.evaluateModelOnce((Classifier)classifier,
test.instance(0));
test.delete(0);
}
testReader.close();
text.append("\n\n" + testingEvaluation.
toSummaryString("=== Error on test data ===\n",
printComplexityStatistics));
} else if (trainFileName.length() != 0) {
// Testing is via cross-validation on training data
random = new Random(seed);
random.setSeed(seed);
train.randomize(random);
testingEvaluation.
crossValidateModel(classifier, train, folds);
if (template.classAttribute().isNumeric()) {
text.append("\n\n\n" + testingEvaluation.
toSummaryString("=== Cross-validation ===\n",
printComplexityStatistics));
} else {
text.append("\n\n\n" + testingEvaluation.
toSummaryString("=== Stratified " +
"cross-validation ===\n",
printComplexityStatistics));
}
}
if (template.classAttribute().isNominal()) {
if (classStatistics) {
text.append("\n\n" + testingEvaluation.toClassDetailsString());
}
text.append("\n\n" + testingEvaluation.toMatrixString());
}
return text.toString();
}
/**
* Attempts to load a cost matrix.
*
* @param costFileName the filename of the cost matrix
* @param numClasses the number of classes that should be in the cost matrix
* (only used if the cost file is in old format).
* @return a <code>CostMatrix</code> value, or null if costFileName is empty
* @exception Exception if an error occurs.
*/
private static CostMatrix handleCostOption(String costFileName,
int numClasses)
throws Exception {
if ((costFileName != null) && (costFileName.length() != 0)) {
System.out.println(
"NOTE: The behaviour of the -m option has changed between WEKA 3.0"
+" and WEKA 3.1. -m now carries out cost-sensitive *evaluation*"
+" only. For cost-sensitive *prediction*, use one of the"
+" cost-sensitive metaschemes such as"
+" weka.classifiers.meta.CostSensitiveClassifier or"
+" weka.classifiers.meta.MetaCost");
Reader costReader = null;
try {
costReader = new BufferedReader(new FileReader(costFileName));
} catch (Exception e) {
throw new Exception("Can't open file " + e.getMessage() + '.');
}
try {
// First try as a proper cost matrix format
return new CostMatrix(costReader);
} catch (Exception ex) {
try {
// Now try as the poxy old format :-)
//System.err.println("Attempting to read old format cost file");
try {
costReader.close(); // Close the old one
costReader = new BufferedReader(new FileReader(costFileName));
} catch (Exception e) {
throw new Exception("Can't open file " + e.getMessage() + '.');
}
CostMatrix costMatrix = new CostMatrix(numClasses);
//System.err.println("Created default cost matrix");
costMatrix.readOldFormat(costReader);
return costMatrix;
//System.err.println("Read old format");
} catch (Exception e2) {
// re-throw the original exception
//System.err.println("Re-throwing original exception");
throw ex;
}
}
} else {
return null;
}
}
/**
* Evaluates the classifier on a given set of instances.
*
* @param classifier machine learning classifier
* @param data set of test instances for evaluation
* @exception Exception if model could not be evaluated
* successfully
*/
public void evaluateModel(Classifier classifier,Instances data)throws Exception{
double [] predicted;
for (int i = 0; i < data.numInstances(); i++) {
evaluateModelOnce((Classifier)classifier,data.instance(i));
}
}
/**
* Evaluates the classifier on a single instance.
*
* @param classifier machine learning classifier
* @param instance the test instance to be classified
* @return the prediction made by the clasifier
* @exception Exception if model could not be evaluated
* successfully or the data contains string attributes
*/
public double evaluateModelOnce(Classifier classifier,
Instance instance) throws Exception {
Instance classMissing = (Instance)instance.copy();
double pred=0;
classMissing.setDataset(instance.dataset());
classMissing.setClassMissing();
if (m_ClassIsNominal) {
if (classifier instanceof DistributionClassifier) {
double [] dist = ((DistributionClassifier)classifier).
distributionForInstance(classMissing);
pred = Utils.maxIndex(dist);
updateStatsForClassifier(dist,
instance);
} else {
pred = classifier.classifyInstance(classMissing);
updateStatsForClassifier(makeDistribution(pred),
instance);
}
} else {
pred = classifier.classifyInstance(classMissing);
updateStatsForPredictor(pred,
instance);
}
return pred;
}
/**
* Evaluates the supplied distribution on a single instance.
*
* @param dist the supplied distribution
* @param instance the test instance to be classified
* @exception Exception if model could not be evaluated
* successfully
*/
public double evaluateModelOnce(double [] dist,
Instance instance) throws Exception {
double pred;
if (m_ClassIsNominal) {
pred = Utils.maxIndex(dist);
updateStatsForClassifier(dist, instance);
} else {
pred = dist[0];
updateStatsForPredictor(pred, instance);
}
return pred;
}
/**
* Evaluates the supplied prediction on a single instance.
*
* @param prediction the supplied prediction
* @param instance the test instance to be classified
* @exception Exception if model could not be evaluated
* successfully
*/
public void evaluateModelOnce(double prediction,
Instance instance) throws Exception {
if (m_ClassIsNominal) {
updateStatsForClassifier(makeDistribution(prediction),
instance);
} else {
updateStatsForPredictor(prediction, instance);
}
}
/**
* Wraps a static classifier in enough source to test using the weka
* class libraries.
*
* @param classifier a Sourcable Classifier
* @param className the name to give to the source code class
* @return the source for a static classifier that can be tested with
* weka libraries.
*/
protected static String wekaStaticWrapper(Sourcable classifier,
String className)
throws Exception {
//String className = "StaticClassifier";
String staticClassifier = classifier.toSource(className);
return "package weka.classifiers;\n"
+"import weka.core.Attribute;\n"
+"import weka.core.Instance;\n"
+"import weka.core.Instances;\n"
+"import weka.classifiers.Classifier;\n\n"
+"public class WekaWrapper extends Classifier {\n\n"
+" public void buildClassifier(Instances i) throws Exception {\n"
+" }\n\n"
+" public double classifyInstance(Instance i) throws Exception {\n\n"
+" Object [] s = new Object [i.numAttributes()];\n"
+" for (int j = 0; j < s.length; j++) {\n"
+" if (!i.isMissing(j)) {\n"
+" if (i.attribute(j).type() == Attribute.NOMINAL) {\n"
+" s[j] = i.attribute(j).value((int) i.value(j));\n"
+" } else if (i.attribute(j).type() == Attribute.NUMERIC) {\n"
+" s[j] = new Double(i.value(j));\n"
+" }\n"
+" }\n"
+" }\n"
+" return " + className + ".classify(s);\n"
+" }\n\n"
+"}\n\n"
+staticClassifier; // The static classifer class
}
/**
* Gets the number of test instances that had a known class value
* (actually the sum of the weights of test instances with known
* class value).
*
* @return the number of test instances with known class
*/
public final double numInstances() {
return m_WithClass;
}
/**
* Gets the number of instances incorrectly classified (that is, for
* which an incorrect prediction was made). (Actually the sum of the weights
* of these instances)
*
* @return the number of incorrectly classified instances
*/
public final double incorrect() {
return m_Incorrect;
}
/**
* Gets the percentage of instances incorrectly classified (that is, for
* which an incorrect prediction was made).
*
* @return the percent of incorrectly classified instances
* (between 0 and 100)
*/
public final double pctIncorrect() {
return 100 * m_Incorrect / m_WithClass;
}
/**
* Gets the total cost, that is, the cost of each prediction times the
* weight of the instance, summed over all instances.
*
* @return the total cost
*/
public final double totalCost() {
return m_TotalCost;
}
/**
* Gets the average cost, that is, total cost of misclassifications
* (incorrect plus unclassified) over the total number of instances.
*
* @return the average cost.
*/
public final double avgCost() {
return m_TotalCost / m_WithClass;
}
/**
* Gets the number of instances correctly classified (that is, for
* which a correct prediction was made). (Actually the sum of the weights
* of these instances)
*
* @return the number of correctly classified instances
*/
public final double correct() {
return m_Correct;
}
/**
* Gets the percentage of instances correctly classified (that is, for
* which a correct prediction was made).
*
* @return the percent of correctly classified instances (between 0 and 100)
*/
public final double pctCorrect() {
return 100 * m_Correct / m_WithClass;
}
/**
* Gets the number of instances not classified (that is, for
* which no prediction was made by the classifier). (Actually the sum
* of the weights of these instances)
*
* @return the number of unclassified instances
*/
public final double unclassified() {
return m_Unclassified;
}
/**
* Gets the percentage of instances not classified (that is, for
* which no prediction was made by the classifier).
*
* @return the percent of unclassified instances (between 0 and 100)
*/
public final double pctUnclassified() {
return 100 * m_Unclassified / m_WithClass;
}
/**
* Returns the estimated error rate or the root mean squared error
* (if the class is numeric). If a cost matrix was given this
* error rate gives the average cost.
*
* @return the estimated error rate (between 0 and 1, or between 0 and
* maximum cost)
*/
public final double errorRate() {
if (!m_ClassIsNominal) {
return Math.sqrt(m_SumSqrErr / m_WithClass);
}
if (m_CostMatrix == null) {
return m_Incorrect / m_WithClass;
} else {
return avgCost();
}
}
/**
* Returns value of kappa statistic if class is nominal.
*
* @return the value of the kappa statistic
*/
public final double kappa() {
double[] sumRows = new double[m_ConfusionMatrix.length];
double[] sumColumns = new double[m_ConfusionMatrix.length];
double sumOfWeights = 0;
for (int i = 0; i < m_ConfusionMatrix.length; i++) {
for (int j = 0; j < m_ConfusionMatrix.length; j++) {
sumRows[i] += m_ConfusionMatrix[i][j];
sumColumns[j] += m_ConfusionMatrix[i][j];
sumOfWeights += m_ConfusionMatrix[i][j];
}
}
double correct = 0, chanceAgreement = 0;
for (int i = 0; i < m_ConfusionMatrix.length; i++) {
chanceAgreement += (sumRows[i] * sumColumns[i]);
correct += m_ConfusionMatrix[i][i];
}
chanceAgreement /= (sumOfWeights * sumOfWeights);
correct /= sumOfWeights;
if (chanceAgreement < 1) {
return (correct - chanceAgreement) / (1 - chanceAgreement);
} else {
return 1;
}
}
/**
* Returns the correlation coefficient if the class is numeric.
*
* @return the correlation coefficient
* @exception Exception if class is not numeric
*/
public final double correlationCoefficient() throws Exception {
if (m_ClassIsNominal) {
throw
new Exception("Can't compute correlation coefficient: " +
"class is nominal!");
}
double correlation = 0;
double varActual =
m_SumSqrClass - m_SumClass * m_SumClass / m_WithClass;
double varPredicted =
m_SumSqrPredicted - m_SumPredicted * m_SumPredicted /
m_WithClass;
double varProd =
m_SumClassPredicted - m_SumClass * m_SumPredicted / m_WithClass;
if (Utils.smOrEq(varActual * varPredicted, 0.0)) {
correlation = 0.0;
} else {
correlation = varProd / Math.sqrt(varActual * varPredicted);
}
return correlation;
}
/**
* Returns the mean absolute error. Refers to the error of the
* predicted values for numeric classes, and the error of the
* predicted probability distribution for nominal classes.
*
* @return the mean absolute error
*/
public final double meanAbsoluteError() {
return m_SumAbsErr / m_WithClass;
}
/**
* Returns the mean absolute error of the prior.
*
* @return the mean absolute error
*/
public final double meanPriorAbsoluteError() {
return m_SumPriorAbsErr / m_WithClass;
}
/**
* Returns the relative absolute error.
*
* @return the relative absolute error
* @exception Exception if it can't be computed
*/
public final double relativeAbsoluteError() throws Exception {
return 100 * meanAbsoluteError() / meanPriorAbsoluteError();
}
/**
* Returns the root mean squared error.
*
* @return the root mean squared error
*/
public final double rootMeanSquaredError() {
return Math.sqrt(m_SumSqrErr / m_WithClass);
}
/**
* Returns the root mean prior squared error.
*
* @return the root mean prior squared error
*/
public final double rootMeanPriorSquaredError() {
return Math.sqrt(m_SumPriorSqrErr / m_WithClass);
}
/**
* Returns the root relative squared error if the class is numeric.
*
* @return the root relative squared error
*/
public final double rootRelativeSquaredError() {
return 100.0 * rootMeanSquaredError() /
rootMeanPriorSquaredError();
}
/**
* Calculate the entropy of the prior distribution
*
* @return the entropy of the prior distribution
* @exception Exception if the class is not nominal
*/
public final double priorEntropy() throws Exception {
if (!m_ClassIsNominal) {
throw
new Exception("Can't compute entropy of class prior: " +
"class numeric!");
}
double entropy = 0;
for(int i = 0; i < m_NumClasses; i++) {
entropy -= m_ClassPriors[i] / m_ClassPriorsSum
* Utils.log2(m_ClassPriors[i] / m_ClassPriorsSum);
}
return entropy;
}
/**
* Return the total Kononenko & Bratko Information score in bits
*
* @return the K&B information score
* @exception Exception if the class is not nominal
*/
public final double KBInformation() throws Exception {
if (!m_ClassIsNominal) {
throw
new Exception("Can't compute K&B Info score: " +
"class numeric!");
}
return m_SumKBInfo;
}
/**
* Return the Kononenko & Bratko Information score in bits per
* instance.
*
* @return the K&B information score
* @exception Exception if the class is not nominal
*/
public final double KBMeanInformation() throws Exception {
if (!m_ClassIsNominal) {
throw
new Exception("Can't compute K&B Info score: "
+ "class numeric!");
}
return m_SumKBInfo / m_WithClass;
}
/**
* Return the Kononenko & Bratko Relative Information score
*
* @return the K&B relative information score
* @exception Exception if the class is not nominal
*/
public final double KBRelativeInformation() throws Exception {
if (!m_ClassIsNominal) {
throw
new Exception("Can't compute K&B Info score: " +
"class numeric!");
}
return 100.0 * KBInformation() / priorEntropy();
}
/**
* Returns the total entropy for the null model
*
* @return the total null model entropy
*/
public final double SFPriorEntropy() {
return m_SumPriorEntropy;
}
/**
* Returns the entropy per instance for the null model
*
* @return the null model entropy per instance
*/
public final double SFMeanPriorEntropy() {
return m_SumPriorEntropy / m_WithClass;
}
/**
* Returns the total entropy for the scheme
*
* @return the total scheme entropy
*/
public final double SFSchemeEntropy() {
return m_SumSchemeEntropy;
}
/**
* Returns the entropy per instance for the scheme
*
* @return the scheme entropy per instance
*/
public final double SFMeanSchemeEntropy() {
return m_SumSchemeEntropy / m_WithClass;
}
/**
* Returns the total SF, which is the null model entropy minus
* the scheme entropy.
*
* @return the total SF
*/
public final double SFEntropyGain() {
return m_SumPriorEntropy - m_SumSchemeEntropy;
}
/**
* Returns the SF per instance, which is the null model entropy
* minus the scheme entropy, per instance.
*
* @return the SF per instance
*/
public final double SFMeanEntropyGain() {
return (m_SumPriorEntropy - m_SumSchemeEntropy) / m_WithClass;
}
/**
* Output the cumulative margin distribution as a string suitable
* for input for gnuplot or similar package.
*
* @return the cumulative margin distribution
* @exception Exception if the class attribute is nominal
*/
public String toCumulativeMarginDistributionString() throws Exception {
if (!m_ClassIsNominal) {
throw new Exception("Class must be nominal for margin distributions");
}
String result = "";
double cumulativeCount = 0;
double margin;
for(int i = 0; i <= k_MarginResolution; i++) {
if (m_MarginCounts[i] != 0) {
cumulativeCount += m_MarginCounts[i];
margin = (double)i * 2.0 / k_MarginResolution - 1.0;
result = result + Utils.doubleToString(margin, 7, 3) + ' '
+ Utils.doubleToString(cumulativeCount * 100
/ m_WithClass, 7, 3) + '\n';
} else if (i == 0) {
result = Utils.doubleToString(-1.0, 7, 3) + ' '
+ Utils.doubleToString(0, 7, 3) + '\n';
}
}
return result;
}
/**
* Calls toSummaryString() with no title and no complexity stats
* @return a summary description of the classifier evaluation
*/
public String toSummaryString() {
return toSummaryString("");
}
public String toSummaryString(String title){
return toSummaryString(title,false);
}
/**
* Calls toSummaryString() with a default title.
*
* @param printComplexityStatistics if true, complexity statistics are
* returned as well
*/
public String toSummaryString(boolean printComplexityStatistics) {
return toSummaryString("=== Summary ===\n", printComplexityStatistics);
}
/**
* Outputs the performance statistics in summary form. Lists
* number (and percentage) of instances classified correctly,
* incorrectly and unclassified. Outputs the total number of
* instances classified, and the number of instances (if any)
* that had no class value provided.
*
* @param title the title for the statistics
* @param printComplexityStatistics if true, complexity statistics are
* returned as well
* @return the summary as a String
*/
public String toSummaryString(String title,
boolean printComplexityStatistics) {
double mae, mad = 0;
StringBuffer text = new StringBuffer();
text.append(title + "\n");
try {
if (m_WithClass > 0) {
if (m_ClassIsNominal) {
text.append("Correctly Classified Instances ");
text.append(Utils.doubleToString(correct(), 12, 4) + " " +
Utils.doubleToString(pctCorrect(),
12, 4) + " %\n");
text.append("Incorrectly Classified Instances ");
text.append(Utils.doubleToString(incorrect(), 12, 4) + " " +
Utils.doubleToString(pctIncorrect(),
12, 4) + " %\n");
text.append("Kappa statistic ");
text.append(Utils.doubleToString(kappa(), 12, 4) + "\n");
if (m_CostMatrix != null) {
text.append("Total Cost ");
text.append(Utils.doubleToString(totalCost(), 12, 4) + "\n");
text.append("Average Cost ");
text.append(Utils.doubleToString(avgCost(), 12, 4) + "\n");
}
if (printComplexityStatistics) {
text.append("K&B Relative Info Score ");
text.append(Utils.doubleToString(KBRelativeInformation(), 12, 4)
+ " %\n");
text.append("K&B Information Score ");
text.append(Utils.doubleToString(KBInformation(), 12, 4)
+ " bits");
text.append(Utils.doubleToString(KBMeanInformation(), 12, 4)
+ " bits/instance\n");
}
} else {
text.append("Correlation coefficient ");
text.append(Utils.doubleToString(correlationCoefficient(), 12 , 4) +
"\n");
}
if (printComplexityStatistics) {
text.append("Class complexity | order 0 ");
text.append(Utils.doubleToString(SFPriorEntropy(), 12, 4)
+ " bits");
text.append(Utils.doubleToString(SFMeanPriorEntropy(), 12, 4)
+ " bits/instance\n");
text.append("Class complexity | scheme ");
text.append(Utils.doubleToString(SFSchemeEntropy(), 12, 4)
+ " bits");
text.append(Utils.doubleToString(SFMeanSchemeEntropy(), 12, 4)
+ " bits/instance\n");
text.append("Complexity improvement (Sf) ");
text.append(Utils.doubleToString(SFEntropyGain(), 12, 4) + " bits");
text.append(Utils.doubleToString(SFMeanEntropyGain(), 12, 4)
+ " bits/instance\n");
}
text.append("Mean absolute error ");
text.append(Utils.doubleToString(meanAbsoluteError(), 12, 4)
+ "\n");
text.append("Root mean squared error ");
text.append(Utils.
doubleToString(rootMeanSquaredError(), 12, 4)
+ "\n");
text.append("Relative absolute error ");
text.append(Utils.doubleToString(relativeAbsoluteError(),
12, 4) + " %\n");
text.append("Root relative squared error ");
text.append(Utils.doubleToString(rootRelativeSquaredError(),
12, 4) + " %\n");
}
if (Utils.gr(unclassified(), 0)) {
text.append("UnClassified Instances ");
text.append(Utils.doubleToString(unclassified(), 12,4) + " " +
Utils.doubleToString(pctUnclassified(),
12, 4) + " %\n");
}
text.append("Total Number of Instances ");
text.append(Utils.doubleToString(m_WithClass, 12, 4) + "\n");
if (m_MissingClass > 0) {
text.append("Ignored Class Unknown Instances ");
text.append(Utils.doubleToString(m_MissingClass, 12, 4) + "\n");
}
} catch (Exception ex) {
// Should never occur since the class is known to be nominal
// here
System.err.println("Arggh - Must be a bug in Evaluation class");
}
return text.toString();
}
/**
* Calls toMatrixString() with a default title.
*
* @return the confusion matrix as a string
* @exception Exception if the class is numeric
*/
public String toMatrixString() throws Exception {
return toMatrixString("=== Confusion Matrix ===\n");
}
/**
* Outputs the performance statistics as a classification confusion
* matrix. For each class value, shows the distribution of
* predicted class values.
*
* @param title the title for the confusion matrix
* @return the confusion matrix as a String
* @exception Exception if the class is numeric
*/
public String toMatrixString(String title) throws Exception {
StringBuffer text = new StringBuffer();
char [] IDChars = {'a','b','c','d','e','f','g','h','i','j',
'k','l','m','n','o','p','q','r','s','t',
'u','v','w','x','y','z'};
int IDWidth;
boolean fractional = false;
if (!m_ClassIsNominal) {
throw new Exception("Evaluation: No confusion matrix possible!");
}
// Find the maximum value in the matrix
// and check for fractional display requirement
double maxval = 0;
for(int i = 0; i < m_NumClasses; i++) {
for(int j = 0; j < m_NumClasses; j++) {
double current = m_ConfusionMatrix[i][j];
if (current < 0) {
current *= -10;
}
if (current > maxval) {
maxval = current;
}
double fract = current - Math.rint(current);
if (!fractional
&& ((Math.log(fract) / Math.log(10)) >= -2)) {
fractional = true;
}
}
}
IDWidth = 1 + Math.max((int)(Math.log(maxval) / Math.log(10)
+ (fractional ? 3 : 0)),
(int)(Math.log(m_NumClasses) /
Math.log(IDChars.length)));
text.append(title).append("\n");
for(int i = 0; i < m_NumClasses; i++) {
if (fractional) {
text.append(" ").append(num2ShortID(i,IDChars,IDWidth - 3))
.append(" ");
} else {
text.append(" ").append(num2ShortID(i,IDChars,IDWidth));
}
}
text.append(" <-- classified as\n");
for(int i = 0; i< m_NumClasses; i++) {
for(int j = 0; j < m_NumClasses; j++) {
text.append(" ").append(
Utils.doubleToString(m_ConfusionMatrix[i][j],
IDWidth,
(fractional ? 2 : 0)));
}
text.append(" | ").append(num2ShortID(i,IDChars,IDWidth))
.append(" = ").append(m_ClassNames[i]).append("\n");
}
return text.toString();
}
public String toClassDetailsString() throws Exception {
return toClassDetailsString("=== Detailed Accuracy By Class ===\n");
}
/**
* Generates a breakdown of the accuracy for each class,
* incorporating various information-retrieval statistics, such as
* true/false positive rate, precision/recall/F-Measure. Should be
* useful for ROC curves, recall/precision curves.
*
* @param title the title to prepend the stats string with
* @return the statistics presented as a string
*/
public String toClassDetailsString(String title) throws Exception {
if (!m_ClassIsNominal) {
throw new Exception("Evaluation: No confusion matrix possible!");
}
StringBuffer text = new StringBuffer(title
+ "\nTP Rate FP Rate"
+ " Precision Recall"
+ " F-Measure Class\n");
for(int i = 0; i < m_NumClasses; i++) {
text.append(Utils.doubleToString(truePositiveRate(i), 7, 3))
.append(" ");
text.append(Utils.doubleToString(falsePositiveRate(i), 7, 3))
.append(" ");
text.append(Utils.doubleToString(precision(i), 7, 3))
.append(" ");
text.append(Utils.doubleToString(recall(i), 7, 3))
.append(" ");
text.append(Utils.doubleToString(fMeasure(i), 7, 3))
.append(" ");
text.append(m_ClassNames[i]).append('\n');
}
return text.toString();
}
/**
* Calculate the number of true positives with respect to a particular class.
* This is defined as<p>
* <pre>
* correctly classified positives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the true positive rate
*/
public double numTruePositives(int classIndex) {
double correct = 0;
for (int j = 0; j < m_NumClasses; j++) {
if (j == classIndex) {
correct += m_ConfusionMatrix[classIndex][j];
}
}
return correct;
}
/**
* Calculate the true positive rate with respect to a particular class.
* This is defined as<p>
* <pre>
* correctly classified positives
* ------------------------------
* total positives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the true positive rate
*/
public double truePositiveRate(int classIndex) {
double correct = 0, total = 0;
for (int j = 0; j < m_NumClasses; j++) {
if (j == classIndex) {
correct += m_ConfusionMatrix[classIndex][j];
}
total += m_ConfusionMatrix[classIndex][j];
}
if (total == 0) {
return 0;
}
return correct / total;
}
/**
* Calculate the number of true negatives with respect to a particular class.
* This is defined as<p>
* <pre>
* correctly classified negatives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the true positive rate
*/
public double numTrueNegatives(int classIndex) {
double correct = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i != classIndex) {
for (int j = 0; j < m_NumClasses; j++) {
if (j != classIndex) {
correct += m_ConfusionMatrix[i][j];
}
}
}
}
return correct;
}
/**
* Calculate the true negative rate with respect to a particular class.
* This is defined as<p>
* <pre>
* correctly classified negatives
* ------------------------------
* total negatives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the true positive rate
*/
public double trueNegativeRate(int classIndex) {
double correct = 0, total = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i != classIndex) {
for (int j = 0; j < m_NumClasses; j++) {
if (j != classIndex) {
correct += m_ConfusionMatrix[i][j];
}
total += m_ConfusionMatrix[i][j];
}
}
}
if (total == 0) {
return 0;
}
return correct / total;
}
/**
* Calculate number of false positives with respect to a particular class.
* This is defined as<p>
* <pre>
* incorrectly classified negatives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the false positive rate
*/
public double numFalsePositives(int classIndex) {
double incorrect = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i != classIndex) {
for (int j = 0; j < m_NumClasses; j++) {
if (j == classIndex) {
incorrect += m_ConfusionMatrix[i][j];
}
}
}
}
return incorrect;
}
/**
* Calculate the false positive rate with respect to a particular class.
* This is defined as<p>
* <pre>
* incorrectly classified negatives
* --------------------------------
* total negatives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the false positive rate
*/
public double falsePositiveRate(int classIndex) {
double incorrect = 0, total = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i != classIndex) {
for (int j = 0; j < m_NumClasses; j++) {
if (j == classIndex) {
incorrect += m_ConfusionMatrix[i][j];
}
total += m_ConfusionMatrix[i][j];
}
}
}
if (total == 0) {
return 0;
}
return incorrect / total;
}
/**
* Calculate number of false negatives with respect to a particular class.
* This is defined as<p>
* <pre>
* incorrectly classified positives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the false positive rate
*/
public double numFalseNegatives(int classIndex) {
double incorrect = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i == classIndex) {
for (int j = 0; j < m_NumClasses; j++) {
if (j != classIndex) {
incorrect += m_ConfusionMatrix[i][j];
}
}
}
}
return incorrect;
}
/**
* Calculate the false negative rate with respect to a particular class.
* This is defined as<p>
* <pre>
* incorrectly classified positives
* --------------------------------
* total positives
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the false positive rate
*/
public double falseNegativeRate(int classIndex) {
double incorrect = 0, total = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i == classIndex) {
for (int j = 0; j < m_NumClasses; j++) {
if (j != classIndex) {
incorrect += m_ConfusionMatrix[i][j];
}
total += m_ConfusionMatrix[i][j];
}
}
}
if (total == 0) {
return 0;
}
return incorrect / total;
}
/**
* Calculate the recall with respect to a particular class.
* This is defined as<p>
* <pre>
* correctly classified positives
* ------------------------------
* total positives
* </pre><p>
* (Which is also the same as the truePositiveRate.)
*
* @param classIndex the index of the class to consider as "positive"
* @return the recall
*/
public double recall(int classIndex) {
return truePositiveRate(classIndex);
}
/**
* Calculate the precision with respect to a particular class.
* This is defined as<p>
* <pre>
* correctly classified positives
* ------------------------------
* total predicted as positive
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the precision
*/
public double precision(int classIndex) {
double correct = 0, total = 0;
for (int i = 0; i < m_NumClasses; i++) {
if (i == classIndex) {
correct += m_ConfusionMatrix[i][classIndex];
}
total += m_ConfusionMatrix[i][classIndex];
}
if (total == 0) {
return 0;
}
return correct / total;
}
/**
* Calculate the F-Measure with respect to a particular class.
* This is defined as<p>
* <pre>
* 2 * recall * precision
* ----------------------
* recall + precision
* </pre>
*
* @param classIndex the index of the class to consider as "positive"
* @return the F-Measure
*/
public double fMeasure(int classIndex) {
double precision = precision(classIndex);
double recall = recall(classIndex);
if ((precision + recall) == 0) {
return 0;
}
return 2 * precision * recall / (precision + recall);
}
/**
* Sets the class prior probabilities
*
* @param train the training instances used to determine
* the prior probabilities
* @exception Exception if the class attribute of the instances is not
* set
*/
public void setPriors(Instances train) throws Exception {
if (!m_ClassIsNominal) {
m_NumTrainClassVals = 0;
m_TrainClassVals = null;
m_TrainClassWeights = null;
m_PriorErrorEstimator = null;
m_ErrorEstimator = null;
for (int i = 0; i < train.numInstances(); i++) {
Instance currentInst = train.instance(i);
if (!currentInst.classIsMissing()) {
addNumericTrainClass(currentInst.classValue(),
currentInst.weight());
}
}
} else {
for (int i = 0; i < m_NumClasses; i++) {
m_ClassPriors[i] = 1;
}
m_ClassPriorsSum = m_NumClasses;
for (int i = 0; i < train.numInstances(); i++) {
if (!train.instance(i).classIsMissing()) {
m_ClassPriors[(int)train.instance(i).classValue()] +=
train.instance(i).weight();
m_ClassPriorsSum += train.instance(i).weight();
}
}
}
}
/**
* Updates the class prior probabilities (when incrementally
* training)
*
* @param instance the new training instance seen
* @exception Exception if the class of the instance is not
* set
*/
public void updatePriors(Instance instance) throws Exception {
if (!instance.classIsMissing()) {
if (!m_ClassIsNominal) {
if (!instance.classIsMissing()) {
addNumericTrainClass(instance.classValue(),
instance.weight());
}
} else {
m_ClassPriors[(int)instance.classValue()] +=
instance.weight();
m_ClassPriorsSum += instance.weight();
}
}
}
/**
* Tests whether the current evaluation object is equal to another
* evaluation object
*
* @param obj the object to compare against
* @return true if the two objects are equal
*/
public boolean equals(Object obj) {
if ((obj == null) || !(obj.getClass().equals(this.getClass()))) {
return false;
}
Evaluation cmp = (Evaluation) obj;
if (m_ClassIsNominal != cmp.m_ClassIsNominal) return false;
if (m_NumClasses != cmp.m_NumClasses) return false;
if (m_Incorrect != cmp.m_Incorrect) return false;
if (m_Correct != cmp.m_Correct) return false;
if (m_Unclassified != cmp.m_Unclassified) return false;
if (m_MissingClass != cmp.m_MissingClass) return false;
if (m_WithClass != cmp.m_WithClass) return false;
if (m_SumErr != cmp.m_SumErr) return false;
if (m_SumAbsErr != cmp.m_SumAbsErr) return false;
if (m_SumSqrErr != cmp.m_SumSqrErr) return false;
if (m_SumClass != cmp.m_SumClass) return false;
if (m_SumSqrClass != cmp.m_SumSqrClass) return false;
if (m_SumPredicted != cmp.m_SumPredicted) return false;
if (m_SumSqrPredicted != cmp.m_SumSqrPredicted) return false;
if (m_SumClassPredicted != cmp.m_SumClassPredicted) return false;
if (m_ClassIsNominal) {
for (int i = 0; i < m_NumClasses; i++) {
for (int j = 0; j < m_NumClasses; j++) {
if (m_ConfusionMatrix[i][j] != cmp.m_ConfusionMatrix[i][j]) {
return false;
}
}
}
}
return true;
}
/**
* Prints the predictions for the given dataset into a String variable.
*/
public String Classifications(Classifier classifier,String testFileName)throws Exception{
return printClassifications(classifier,template,testFileName,-1,null);
}
public String Classifications(Classifier classifier,FileReader testReader)throws Exception{
return printClassifications(classifier,template,testReader,-1,null);
}
private static String printClassifications(Classifier classifier,Instances train,String testFileName,int classIndex,Range attributesToOutput)throws Exception{
if(testFileName.length()==0)return "";
FileReader testReader=null;
try{
testReader=new FileReader(testFileName);
}catch(Exception e){
throw new Exception("Can't open file "+e.getMessage()+".");
}
return printClassifications(classifier,train,testReader,classIndex,attributesToOutput);
}
private static String printClassifications(Classifier classifier,
Instances train,
FileReader fileReader,
int classIndex,
Range attributesToOutput) throws Exception {
StringBuffer text = new StringBuffer();
BufferedReader testReader=new BufferedReader(fileReader);
Instances test=new Instances(testReader,1);
if(classIndex!=-1)test.setClassIndex(classIndex-1);
else test.setClassIndex(-1);
int i=0;
while(test.readInstance(testReader)){
Instance instance=test.instance(0);
Instance withMissing=(Instance)instance.copy();
withMissing.setDataset(test);
double predValue=((Classifier)classifier).classifyInstance(withMissing);
if(test.classAttribute().isNumeric()){
if(Instance.isMissingValue(predValue)){
text.append(i+" missing ");
}else{
text.append(i+" "+predValue+" ");
}
if(instance.classIsMissing()){
text.append("missing");
}else{
text.append(instance.classValue());
}
text.append(" "+attributeValuesString(withMissing, attributesToOutput)+"\n");
}else{
if(Instance.isMissingValue(predValue)){
text.append(i+" missing ");
}else{
text.append(i+" "+test.classAttribute().value((int)predValue)+" ");
}
if(classifier instanceof DistributionClassifier){
if(Instance.isMissingValue(predValue)){
text.append("missing ");
}else{
text.append(((DistributionClassifier)classifier).distributionForInstance(withMissing)[(int)predValue]+" ");
}
}
text.append(instance.toString(instance.classIndex())+" "+attributeValuesString(withMissing,attributesToOutput)+"\n");
}
test.delete(0);
i++;
}
testReader.close();
return text.toString();
}
/**
* Builds a string listing the attribute values in a specified range of indices,
* separated by commas and enclosed in brackets.
*
* @param instance the instance to print the values from
* @param attributes the range of the attributes to list
* @return a string listing values of the attributes in the range
*/
private static String attributeValuesString(Instance instance, Range attRange) {
StringBuffer text = new StringBuffer();
if (attRange != null) {
boolean firstOutput = true;
attRange.setUpper(instance.numAttributes() - 1);
for (int i=0; i<instance.numAttributes(); i++)
if (attRange.isInRange(i) && i != instance.classIndex()) {
if (firstOutput) text.append("(");
else text.append(",");
text.append(instance.toString(i));
firstOutput = false;
}
if (!firstOutput) text.append(")");
}
return text.toString();
}
/**
* Make up the help string giving all the command line options
*
* @param classifier the classifier to include options for
* @return a string detailing the valid command line options
*/
private static String makeOptionString(Classifier classifier) {
StringBuffer optionsText = new StringBuffer("");
// General options
optionsText.append("\n\nGeneral options:\n\n");
optionsText.append("-t <name of training file>\n");
optionsText.append("\tSets training file.\n");
optionsText.append("-T <name of test file>\n");
optionsText.append("\tSets test file. If missing, a cross-validation");
optionsText.append(" will be performed on the training data.\n");
optionsText.append("-c <class index>\n");
optionsText.append("\tSets index of class attribute (default: last).\n");
optionsText.append("-x <number of folds>\n");
optionsText.append("\tSets number of folds for cross-validation (default: 10).\n");
optionsText.append("-s <random number seed>\n");
optionsText.append("\tSets random number seed for cross-validation (default: 1).\n");
optionsText.append("-m <name of file with cost matrix>\n");
optionsText.append("\tSets file with cost matrix.\n");
optionsText.append("-l <name of input file>\n");
optionsText.append("\tSets model input file.\n");
optionsText.append("-d <name of output file>\n");
optionsText.append("\tSets model output file.\n");
optionsText.append("-v\n");
optionsText.append("\tOutputs no statistics for training data.\n");
optionsText.append("-o\n");
optionsText.append("\tOutputs statistics only, not the classifier.\n");
optionsText.append("-i\n");
optionsText.append("\tOutputs detailed information-retrieval");
optionsText.append(" statistics for each class.\n");
optionsText.append("-k\n");
optionsText.append("\tOutputs information-theoretic statistics.\n");
optionsText.append("-p <attribute range>\n");
optionsText.append("\tOnly outputs predictions for test instances, along with attributes "
+ "(0 for none).\n");
optionsText.append("-r\n");
optionsText.append("\tOnly outputs cumulative margin distribution.\n");
if (classifier instanceof Sourcable) {
optionsText.append("-z <class name>\n");
optionsText.append("\tOnly outputs the source representation"
+ " of the classifier, giving it the supplied"
+ " name.\n");
}
if (classifier instanceof Drawable) {
optionsText.append("-g\n");
optionsText.append("\tOnly outputs the graph representation"
+ " of the classifier.\n");
}
// Get scheme-specific options
if (classifier instanceof OptionHandler) {
optionsText.append("\nOptions specific to "
+ classifier.getClass().getName()
+ ":\n\n");
Enumeration enum = ((OptionHandler)classifier).listOptions();
while (enum.hasMoreElements()) {
Option option = (Option) enum.nextElement();
optionsText.append(option.synopsis() + '\n');
optionsText.append(option.description() + "\n");
}
}
return optionsText.toString();
}
/**
* Method for generating indices for the confusion matrix.
*
* @param num integer to format
* @return the formatted integer as a string
*/
private String num2ShortID(int num,char [] IDChars,int IDWidth) {
char ID [] = new char [IDWidth];
int i;
for(i = IDWidth - 1; i >=0; i--) {
ID[i] = IDChars[num % IDChars.length];
num = num / IDChars.length - 1;
if (num < 0) {
break;
}
}
for(i--; i >= 0; i--) {
ID[i] = ' ';
}
return new String(ID);
}
/**
* Convert a single prediction into a probability distribution
* with all zero probabilities except the predicted value which
* has probability 1.0;
*
* @param predictedClass the index of the predicted class
* @return the probability distribution
*/
private double [] makeDistribution(double predictedClass) {
double [] result = new double [m_NumClasses];
if (Instance.isMissingValue(predictedClass)) {
return result;
}
if (m_ClassIsNominal) {
result[(int)predictedClass] = 1.0;
} else {
result[0] = predictedClass;
}
return result;
}
/**
* Updates all the statistics about a classifiers performance for
* the current test instance.
*
* @param predictedDistribution the probabilities assigned to
* each class
* @param instance the instance to be classified
* @exception Exception if the class of the instance is not
* set
*/
private void updateStatsForClassifier(double [] predictedDistribution,
Instance instance)
throws Exception {
int actualClass = (int)instance.classValue();
double costFactor = 1;
if (!instance.classIsMissing()) {
updateMargins(predictedDistribution, actualClass, instance.weight());
// Determine the predicted class (doesn't detect multiple
// classifications)
int predictedClass = -1;
double bestProb = 0.0;
for(int i = 0; i < m_NumClasses; i++) {
if (predictedDistribution[i] > bestProb) {
predictedClass = i;
bestProb = predictedDistribution[i];
}
}
m_WithClass += instance.weight();
// Determine misclassification cost
if (m_CostMatrix != null) {
if (predictedClass < 0) {
// For missing predictions, we assume the worst possible cost.
// This is pretty harsh.
// Perhaps we could take the negative of the cost of a correct
// prediction (-m_CostMatrix.getElement(actualClass,actualClass)),
// although often this will be zero
m_TotalCost += instance.weight()
* m_CostMatrix.getMaxCost(actualClass);
} else {
m_TotalCost += instance.weight()
* m_CostMatrix.getElement(actualClass, predictedClass);
}
}
// Update counts when no class was predicted
if (predictedClass < 0) {
m_Unclassified += instance.weight();
return;
}
double predictedProb = Math.max(MIN_SF_PROB,
predictedDistribution[actualClass]);
double priorProb = Math.max(MIN_SF_PROB,
m_ClassPriors[actualClass]
/ m_ClassPriorsSum);
if (predictedProb >= priorProb) {
m_SumKBInfo += (Utils.log2(predictedProb) -
Utils.log2(priorProb))
* instance.weight();
} else {
m_SumKBInfo -= (Utils.log2(1.0-predictedProb) -
Utils.log2(1.0-priorProb))
* instance.weight();
}
m_SumSchemeEntropy -= Utils.log2(predictedProb) * instance.weight();
m_SumPriorEntropy -= Utils.log2(priorProb) * instance.weight();
updateNumericScores(predictedDistribution,
makeDistribution(instance.classValue()),
instance.weight());
// Update other stats
m_ConfusionMatrix[actualClass][predictedClass] += instance.weight();
if (predictedClass != actualClass) {
m_Incorrect += instance.weight();
} else {
m_Correct += instance.weight();
}
} else {
m_MissingClass += instance.weight();
}
}
/**
* Updates all the statistics about a predictors performance for
* the current test instance.
*
* @param predictedValue the numeric value the classifier predicts
* @param instance the instance to be classified
* @exception Exception if the class of the instance is not
* set
*/
private void updateStatsForPredictor(double predictedValue,
Instance instance)
throws Exception {
if (!instance.classIsMissing()){
// Update stats
m_WithClass += instance.weight();
if (Instance.isMissingValue(predictedValue)) {
m_Unclassified += instance.weight();
return;
}
m_SumClass += instance.weight() * instance.classValue();
m_SumSqrClass += instance.weight() * instance.classValue()
* instance.classValue();
m_SumClassPredicted += instance.weight()
* instance.classValue() * predictedValue;
m_SumPredicted += predictedValue;
m_SumSqrPredicted += predictedValue * predictedValue;
if (m_ErrorEstimator == null) {
setNumericPriorsFromBuffer();
}
double predictedProb = Math.max(m_ErrorEstimator.getProbability(
predictedValue
- instance.classValue()),
MIN_SF_PROB);
double priorProb = Math.max(m_PriorErrorEstimator.getProbability(
instance.classValue()),
MIN_SF_PROB);
m_SumSchemeEntropy -= Utils.log2(predictedProb) * instance.weight();
m_SumPriorEntropy -= Utils.log2(priorProb) * instance.weight();
m_ErrorEstimator.addValue(predictedValue - instance.classValue(),
instance.weight());
updateNumericScores(makeDistribution(predictedValue),
makeDistribution(instance.classValue()),
instance.weight());
} else
m_MissingClass += instance.weight();
}
/**
* Update the cumulative record of classification margins
*
* @param predictedDistribution the probability distribution predicted for
* the current instance
* @param actualClass the index of the actual instance class
* @param weight the weight assigned to the instance
*/
private void updateMargins(double [] predictedDistribution,
int actualClass, double weight) {
double probActual = predictedDistribution[actualClass];
double probNext = 0;
for(int i = 0; i < m_NumClasses; i++)
if ((i != actualClass) &&
(predictedDistribution[i] > probNext))
probNext = predictedDistribution[i];
double margin = probActual - probNext;
int bin = (int)((margin + 1.0) / 2.0 * k_MarginResolution);
m_MarginCounts[bin] += weight;
}
/**
* Update the numeric accuracy measures. For numeric classes, the
* accuracy is between the actual and predicted class values. For
* nominal classes, the accuracy is between the actual and
* predicted class probabilities.
*
* @param predicted the predicted values
* @param actual the actual value
* @param weight the weight associated with this prediction
*/
private void updateNumericScores(double [] predicted,
double [] actual, double weight) {
double diff;
double sumErr = 0, sumAbsErr = 0, sumSqrErr = 0;
double sumPriorAbsErr = 0, sumPriorSqrErr = 0;
for(int i = 0; i < m_NumClasses; i++) {
diff = predicted[i] - actual[i];
sumErr += diff;
sumAbsErr += Math.abs(diff);
sumSqrErr += diff * diff;
diff = (m_ClassPriors[i] / m_ClassPriorsSum) - actual[i];
sumPriorAbsErr += Math.abs(diff);
sumPriorSqrErr += diff * diff;
}
m_SumErr += weight * sumErr / m_NumClasses;
m_SumAbsErr += weight * sumAbsErr / m_NumClasses;
m_SumSqrErr += weight * sumSqrErr / m_NumClasses;
m_SumPriorAbsErr += weight * sumPriorAbsErr / m_NumClasses;
m_SumPriorSqrErr += weight * sumPriorSqrErr / m_NumClasses;
}
/**
* Adds a numeric (non-missing) training class value and weight to
* the buffer of stored values.
*
* @param classValue the class value
* @param weight the instance weight
*/
private void addNumericTrainClass(double classValue, double weight) {
if (m_TrainClassVals == null) {
m_TrainClassVals = new double [100];
m_TrainClassWeights = new double [100];
}
if (m_NumTrainClassVals == m_TrainClassVals.length) {
double [] temp = new double [m_TrainClassVals.length * 2];
System.arraycopy(m_TrainClassVals, 0,
temp, 0, m_TrainClassVals.length);
m_TrainClassVals = temp;
temp = new double [m_TrainClassWeights.length * 2];
System.arraycopy(m_TrainClassWeights, 0,
temp, 0, m_TrainClassWeights.length);
m_TrainClassWeights = temp;
}
m_TrainClassVals[m_NumTrainClassVals] = classValue;
m_TrainClassWeights[m_NumTrainClassVals] = weight;
m_NumTrainClassVals++;
}
/**
* Sets up the priors for numeric class attributes from the
* training class values that have been seen so far.
*/
private void setNumericPriorsFromBuffer() {
double numPrecision = 0.01; // Default value
if (m_NumTrainClassVals > 1) {
double [] temp = new double [m_NumTrainClassVals];
System.arraycopy(m_TrainClassVals, 0, temp, 0, m_NumTrainClassVals);
int [] index = Utils.sort(temp);
double lastVal = temp[index[0]];
double currentVal, deltaSum = 0;
int distinct = 0;
for (int i = 1; i < temp.length; i++) {
double current = temp[index[i]];
if (current != lastVal) {
deltaSum += current - lastVal;
lastVal = current;
distinct++;
}
}
if (distinct > 0) {
numPrecision = deltaSum / distinct;
}
}
m_PriorErrorEstimator = new KernelEstimator(numPrecision);
m_ErrorEstimator = new KernelEstimator(numPrecision);
m_ClassPriors[0] = m_ClassPriorsSum = 0.0001; // zf correction
for (int i = 0; i < m_NumTrainClassVals; i++) {
m_ClassPriors[0] += m_TrainClassVals[i] * m_TrainClassWeights[i];
m_ClassPriorsSum += m_TrainClassWeights[i];
m_PriorErrorEstimator.addValue(m_TrainClassVals[i],
m_TrainClassWeights[i]);
}
}
}