public boolean predict(SingleDecision decision) throws MaltChainedException {
if (getGuide().getGuideMode() == ClassifierGuide.GuideMode.BATCH) {
throw new GuideException("Can only predict during parsing. ");
} else if (!(divideFeature.getFeatureValue() instanceof SingleFeatureValue)) {
throw new GuideException("The divide feature does not have a single value. ");
}
// divideFeature.update();
if (divideModels != null
&& divideModels.containsKey(
((SingleFeatureValue) divideFeature.getFeatureValue()).getCode())) {
return divideModels
.get(((SingleFeatureValue) divideFeature.getFeatureValue()).getCode())
.predict(decision);
} else if (masterModel != null && masterModel.getFrequency() > 0) {
return masterModel.predict(decision);
} else {
getGuide()
.getConfiguration()
.getConfigLogger()
.info(
"Could not predict the next parser decision because there is "
+ "no divide or master model that covers the divide value '"
+ ((SingleFeatureValue) divideFeature.getFeatureValue()).getCode()
+ "', as default"
+ " class code '1' is used. ");
decision.addDecision(1); // default prediction
// classCodeTable.getEmptyKBestList().addKBestItem(1);
}
return true;
}
public void addInstance(SingleDecision decision, FeatureVector featureVector)
throws MaltChainedException {
if (featureVector == null) {
throw new LibException("The feature vector cannot be found");
} else if (decision == null) {
throw new LibException("The decision cannot be found");
}
try {
sb.append(decision.getDecisionCode() + "\t");
final int n = featureVector.size();
for (int i = 0; i < n; i++) {
FeatureValue featureValue = featureVector.getFeatureValue(i);
if (featureValue == null || (excludeNullValues == true && featureValue.isNullValue())) {
sb.append("-1");
} else {
if (featureValue instanceof SingleFeatureValue) {
SingleFeatureValue singleFeatureValue = (SingleFeatureValue) featureValue;
if (singleFeatureValue.getValue() == 1) {
sb.append(singleFeatureValue.getIndexCode());
} else if (singleFeatureValue.getValue() == 0) {
sb.append("-1");
} else {
sb.append(singleFeatureValue.getIndexCode());
sb.append(":");
sb.append(singleFeatureValue.getValue());
}
} else if (featureValue instanceof MultipleFeatureValue) {
Set<Integer> values = ((MultipleFeatureValue) featureValue).getCodes();
int j = 0;
for (Integer value : values) {
sb.append(value.toString());
if (j != values.size() - 1) {
sb.append("|");
}
j++;
}
} else {
throw new LibException(
"Don't recognize the type of feature value: " + featureValue.getClass());
}
}
sb.append('\t');
}
sb.append('\n');
instanceOutput.write(sb.toString());
instanceOutput.flush();
increaseNumberOfInstances();
sb.setLength(0);
} catch (IOException e) {
throw new LibException("The learner cannot write to the instance file. ", e);
}
}
/**
* Updates the corresponding single decision with the next value in the k-best list.
*
* @return true if decision has been updated, otherwise false
* @throws MaltChainedException
*/
public boolean updateActionWithNextKBest() throws MaltChainedException {
if (addCandidateIndex != 0
&& topCandidateIndex < addCandidateIndex
&& topCandidateIndex < kBestList.size()) {
int actionCode = kBestList.get(topCandidateIndex).getActionCode();
if (decision instanceof SingleDecision) {
((SingleDecision) decision).addDecision(actionCode);
}
topCandidateIndex++;
return true;
}
return false;
}
/**
* Adds a candidate to the k-best list
*
* @param actionCode the integer representation of candidate action
* @throws MaltChainedException
*/
public void add(int actionCode) throws MaltChainedException {
if (k != -1 && addCandidateIndex >= k) {
return;
}
if (addCandidateIndex >= kBestList.size()) {
kBestList.add(new Candidate());
}
kBestList.get(addCandidateIndex).setActionCode(actionCode);
if (addCandidateIndex == 0) {
if (decision instanceof SingleDecision) {
((SingleDecision) decision).addDecision(actionCode);
}
topCandidateIndex++;
}
addCandidateIndex++;
}
public boolean predict(FeatureVector featureVector, SingleDecision decision)
throws MaltChainedException {
if (featureVector == null) {
throw new LibException(
"The learner cannot predict the next class, because the feature vector cannot be found. ");
}
final FeatureList featureList = new FeatureList();
final int size = featureVector.size();
for (int i = 1; i <= size; i++) {
final FeatureValue featureValue = featureVector.getFeatureValue(i - 1);
if (featureValue != null && !(excludeNullValues == true && featureValue.isNullValue())) {
if (featureValue instanceof SingleFeatureValue) {
SingleFeatureValue singleFeatureValue = (SingleFeatureValue) featureValue;
int index = featureMap.getIndex(i, singleFeatureValue.getIndexCode());
if (index != -1 && singleFeatureValue.getValue() != 0) {
featureList.add(index, singleFeatureValue.getValue());
}
} else if (featureValue instanceof MultipleFeatureValue) {
for (Integer value : ((MultipleFeatureValue) featureValue).getCodes()) {
int v = featureMap.getIndex(i, value);
if (v != -1) {
featureList.add(v, 1);
}
}
}
}
}
try {
decision.getKBestList().addList(model.predict(featureList.toArray()));
// decision.getKBestList().addList(prediction(featureList));
} catch (OutOfMemoryError e) {
throw new LibException("Out of memory. Please increase the Java heap size (-Xmx<size>). ", e);
}
return true;
}