/**
* Trains the classifier on the array of Signal objects. Implementations of this method should
* also produce an ordered list of the class names which can be returned with the <code>
* getClassNames</code> method.
*
* @param inputData the Signal array that the model should be trained on.
* @throws noMetadataException Thrown if there is no class metadata to train the Gaussian model
* with
*/
public void train(Signal[] inputData) {
List classNamesList = new ArrayList();
for (int i = 0; i < inputData.length; i++) {
try {
String className = inputData[i].getStringMetadata(Signal.PROP_CLASS);
if ((className != null) && (!classNamesList.contains(className))) {
classNamesList.add(className);
}
} catch (noMetadataException ex) {
throw new RuntimeException("No class metadata found to train model on!", ex);
}
}
Collections.sort(classNamesList);
classnames = (String[]) classNamesList.toArray(new String[classNamesList.size()]);
FastVector classValues = new FastVector(classnames.length);
for (int i = 0; i < classnames.length; i++) {
classValues.addElement(classnames[i]);
}
classAttribute = new Attribute(Signal.PROP_CLASS, classValues);
Instances trainingDataSet =
new Instances(Signal2Instance.convert(inputData[0], classAttribute));
if (inputData.length > 1) {
for (int i = 1; i < inputData.length; i++) {
Instances aSignalInstance = Signal2Instance.convert(inputData[i], classAttribute);
for (int j = 0; j < aSignalInstance.numInstances(); j++)
trainingDataSet.add(aSignalInstance.instance(j));
}
}
trainingDataSet.setClass(classAttribute);
inputData = null;
theRule = new MISMO();
// parse options
StringTokenizer stOption = new StringTokenizer(this.MISMOOptions, " ");
String[] options = new String[stOption.countTokens()];
for (int i = 0; i < options.length; i++) {
options[i] = stOption.nextToken();
}
try {
theRule.setOptions(options);
} catch (Exception ex) {
throw new RuntimeException("Failed to set MISMO classifier options!", ex);
}
try {
theRule.buildClassifier(trainingDataSet);
System.out.println("WEKA: outputting MISMO classifier; " + theRule.globalInfo());
} catch (Exception ex) {
throw new RuntimeException("Failed to train classifier!", ex);
}
}
/**
* Calculates the probability of class membership of a Signal Object.
*
* @param inputSignal The Signal object to calculate the probabilities of class membership for.
* This probabilities should be ordered such that the indexes match the class names returned
* by <code>getClassNames</code>.
* @return An array of the probabilities of class membership
*/
public double[] probabilities(Signal inputSignal) {
Instances theData = Signal2Instance.convert(inputSignal, this.classAttribute);
double[] probsAccum = new double[this.getNumClasses()];
for (int i = 0; i < theData.numInstances(); i++) {
double[] temp;
try {
temp = theRule.distributionForInstance(theData.instance(i));
// if(this.verbose){
// System.out.print("distribution:" );
// for (int j=0;j<this.getNumClasses();j++) {
// System.out.print(" " + temp[j]);
// }
// System.out.println("");
// }
} catch (Exception ex) {
System.out.println(
"Exception occured when calculating distribution for a vector!\n" + ex.getMessage());
ex.printStackTrace();
return null;
}
for (int j = 0; j < this.getNumClasses(); j++) {
probsAccum[j] += Math.log(temp[j]);
}
}
// if(this.verbose){
// System.out.print("Log Probs:" );
// for (int j=0;j<this.getNumClasses();j++) {
// System.out.print(" " + probsAccum[j]);
// }
// System.out.println("");
// }
// normalise to range 0:1
double min = Double.POSITIVE_INFINITY;
double max = Double.NEGATIVE_INFINITY;
for (int j = 0; j < this.getNumClasses(); j++) {
if (probsAccum[j] > max) {
max = probsAccum[j];
}
if (probsAccum[j] < min) {
min = probsAccum[j];
}
}
for (int j = 0; j < this.getNumClasses(); j++) {
probsAccum[j] -= min;
probsAccum[j] /= (max - min);
}
// normalise to sum to 1
double total = 0.0;
for (int j = 0; j < this.getNumClasses(); j++) {
total += probsAccum[j];
}
if (verbose) {
DecimalFormat dec = new DecimalFormat();
dec.setMaximumFractionDigits(3);
if (this.verbose) {
System.out.print("Probabilities:");
}
for (int j = 0; j < this.getNumClasses(); j++) {
probsAccum[j] /= total;
if (verbose) {
System.out.print(" " + dec.format(probsAccum[j]));
}
}
if (verbose) {
try {
System.out.println("\t" + inputSignal.getStringMetadata(Signal.PROP_FILE_LOCATION));
} catch (noMetadataException nme) {
nme.printStackTrace(System.out);
}
}
}
return probsAccum;
}