/**
* Populate the FeatureVector with Bag of Words.
*
* @param c
* @param fv
*/
protected void populateFV(String text, FeatureVector<E> fv) {
List<String> unnormalized = tokenizer.tokenize(text);
Multiset<String> terms = HashMultiset.create();
for (String token : unnormalized) {
String norm = Util.normalize(token);
if (!norm.isEmpty()) {
terms.add(norm);
}
}
// sparse representation... no need to put in 0's
for (String term : terms.elementSet()) {
// rare words don't get included, so check first
if (!integerFeatureNames && train.getMetadata().containsKey(term)) {
DoubleFeature bagFeat = new DoubleFeature(term, (double) terms.count(term));
fv.put(term, bagFeat);
} else if (integerFeatureNames
&& train.getMetadata().containsKey(String.valueOf(wordIndexMap.get(term)))) {
String featureName = String.valueOf(wordIndexMap.get(term));
DoubleFeature bagFeat = new DoubleFeature(featureName, (double) terms.count(term));
fv.put(featureName, bagFeat);
}
}
}
/**
* Get the training data.
*
* @return
*/
public TrainRelation<E> getTrainingData() {
TrainRelation<E> copy =
new TrainRelation<E>(train.getName(), (Metadata) train.getMetadata().clone());
copy.addAll(train);
return copy;
}
/**
* Get the training data.
*
* @return
*/
public TrainRelation<E> getTrainingData(E clazz) {
TrainRelation<E> subRelation =
new TrainRelation<E>("sub-relation", (Metadata) train.getMetadata().clone());
for (LabeledFeatureVector<E> lfv : train) {
if (lfv.getLabel().equals(clazz)) {
subRelation.add(lfv);
}
}
return subRelation;
}
/**
* Train the classifier with this instance.
*
* @param c
* @param clazz
*/
public void train(String id, String text, E clazz) {
if (!trained.contains(id)) {
LabeledFeatureVector<E> lfv = createLabeledFV(id, text, clazz);
train.add(lfv);
trained.add(id);
}
}
/**
* Train the classifier with this instance.
*
* @param fv
* @param clazz
*/
public void train(FeatureVector<E> fv, E clazz) {
if (!trained.contains(fv.getId())) {
LabeledFeatureVector<E> lfv = new LabeledFeatureVector<E>(clazz, fv.getId());
lfv.putAll(fv);
lfv.setQid(fv.getQid());
lfv.setRank(fv.getRank());
train.add(lfv);
trained.add(fv.getId());
}
}
/**
* Create the Bag of Words features.
*
* @param citations
*/
public void createFeatures(Collection<String> documents) {
Multiset<String> terms = HashMultiset.create();
for (String s : documents) {
List<String> unnormalized = tokenizer.tokenize(s);
// normalize them
for (int i = 0; i < unnormalized.size(); i++) {
String u = unnormalized.get(i);
String norm = Util.normalize(u);
if (!norm.isEmpty()) {
terms.add(norm);
}
if (bigrams && (i < unnormalized.size() - 1)) {
String second = unnormalized.get(i + 1);
String normSecond = Util.normalize(second);
if (!normSecond.isEmpty()) {
terms.add(norm + "_" + normSecond);
}
}
}
}
int i = 0;
for (String term : terms.elementSet()) {
if (terms.count(term) >= minOccurs // don't count infreq. words
&& term.length() >= minLength) { // or super short words
if (!integerFeatureNames) {
train.getMetadata().put(term, "boolean");
} else {
wordIndexMap.put(term, i++);
train.getMetadata().put(String.valueOf(i), "boolean");
}
}
}
}
/** Clear the training set. */
public void clearTrainingSet() {
train.clear();
trained.clear();
}