@Override
public PredictedLabels predictForTest(
Example vector, Vector W, String realClass, ClassifierData classifierData, int returnAll) {
if (returnAll != Consts.ERROR_NUMBER) {
try {
PredictedLabels tree = new PredictedLabels();
// validation
if (vector.sizeOfVector <= 0) {
Logger.error(ErrorConstants.PHI_VECTOR_DATA);
return null;
}
for (int i = 0; i < numOfClass; i++) {
Example phiData =
classifierData.phi.convert(vector, String.valueOf(i), classifierData.kernel);
// multiple the vectors
double tmp = MathHelpers.multipleVectors(W, phiData.getFeatures());
// get the max value for the max classification
tree.put(String.valueOf(i), tmp);
}
MapValueComparatorDescending vc = new MapValueComparatorDescending(tree);
PredictedLabels result = new PredictedLabels(vc);
result.putAll(tree);
return result;
} catch (Exception e) {
e.printStackTrace();
return null;
}
} else return predictForTrain(vector, W, realClass, classifierData, 0);
}
@Override
// predict function
// argmax(yS,yE) (W*Phi(Xi,yS,yE)) + Task Loss
// this function assumes that the argument vector has already been converted to phi vector
// return null on error
public PredictedLabels predictForTrain(
Example vector,
Vector W,
String realClass,
ClassifierData classifierData,
double epsilonArgMax) {
try {
// validation
if (vector.sizeOfVector <= 0) {
Logger.error(ErrorConstants.PHI_VECTOR_DATA);
return null;
}
if (this.numOfClass > 0) {
// get score for the first label
String maxLabel = "0";
Example phiData =
classifierData.phi.convert(vector, String.valueOf(0), classifierData.kernel);
double maxScore = MathHelpers.multipleVectors(W, phiData.getFeatures());
if (epsilonArgMax != 0) {
// add the task loss
maxScore +=
epsilonArgMax
* classifierData.taskLoss.computeTaskLoss(
String.valueOf(0), realClass, classifierData.arguments);
}
for (int i = 1; i < numOfClass; i++) {
phiData = classifierData.phi.convert(vector, String.valueOf(i), classifierData.kernel);
// multiple the vectors
double tmp = MathHelpers.multipleVectors(W, phiData.getFeatures());
if (epsilonArgMax != 0) {
// add the task loss
tmp +=
epsilonArgMax
* classifierData.taskLoss.computeTaskLoss(
String.valueOf(i), realClass, classifierData.arguments);
}
// updates the max score and max label
if (tmp > maxScore) {
maxLabel = String.valueOf(i);
maxScore = tmp;
}
}
PredictedLabels result = new PredictedLabels();
result.put(maxLabel, maxScore);
return result;
}
return null;
} catch (Exception e) {
e.printStackTrace();
return null;
}
}
