/* (non-Javadoc)
* @see lexelt.ILexeltWriter#getInstances(lexelt.ILexelt)
*/
public Object getInstances(ILexelt p_Lexelt) throws ClassNotFoundException {
svm_problem retVal = new svm_problem();
ArrayList<svm_node[]> featureVectors = new ArrayList<svm_node[]>();
ArrayList<Double> classes = new ArrayList<Double>();
int[][] indice = this.loadStatistic(p_Lexelt);
if (indice == null) {
throw new IllegalArgumentException("the input lexelt should not be null.");
}
IStatistic stat = p_Lexelt.getStatistic();
int size = p_Lexelt.size(); // instance count
for (int i = 0; i < size; i++) {
IInstance instance = p_Lexelt.getInstance(i);
svm_node[] featureVector = this.getVector(instance, stat, indice);
ArrayList<String> tags = instance.getTag();
if (tags.size() > 0) {
for (String tag : tags) {
double c = Double.parseDouble(tag);
featureVectors.add(Arrays.copyOf(featureVector, featureVector.length));
classes.add(c);
}
} else {
featureVectors.add(featureVector);
classes.add(new Double(0));
}
}
retVal.l = featureVectors.size();
retVal.x = new svm_node[retVal.l][];
retVal.y = new double[retVal.l];
for (int i = 0; i < featureVectors.size(); i++) {
retVal.x[i] = featureVectors.get(i);
retVal.y[i] = classes.get(i);
}
return retVal;
}
/**
* get the vector of one instance
*
* @param p_Instance
* @param p_Stat
* @return
*/
private svm_node[] getVector(IInstance p_Instance, IStatistic p_Stat, int[][] p_Indice) {
String value = null;
int kIndex = 0;
int featureSize = p_Instance.size();
Hashtable<Integer, Double> exist = new Hashtable<Integer, Double>();
for (int fIndex = 0; fIndex < featureSize; fIndex++) {
IFeature feature = p_Instance.getFeature(fIndex);
kIndex = p_Stat.getIndex(feature.getKey());
if (kIndex < 0) {
continue;
}
if (ANumericFeature.class.isInstance(feature)) {
exist.put(p_Indice[kIndex][0], Double.parseDouble(feature.getValue()));
} else if (ABinaryFeature.class.isInstance(feature)) {
if (feature.getValue().equals("1")) {
exist.put(p_Indice[kIndex][0], 1.0);
}
} else {
List<String> values = p_Stat.getValue(kIndex);
value = feature.getValue();
if (value == null || !p_Stat.contains(kIndex, value)) {
value = p_Stat.getDefaultValue();
}
for (int i = 0; i < values.size(); i++) {
if (values.get(i).equals(value)) {
exist.put(p_Indice[kIndex][i], 1.0);
break;
}
}
}
}
ArrayList<Integer> indice = new ArrayList<Integer>(exist.keySet());
Collections.sort(indice);
svm_node[] retVal = new svm_node[exist.size()];
for (int i = 0; i < indice.size(); i++) {
retVal[i] = new svm_node();
retVal[i].index = indice.get(i);
retVal[i].value = exist.get(indice.get(i));
}
return retVal;
}
/**
* @param p_Instance
* @param p_Stat
* @return
*/
private String toString(IInstance p_Instance, IStatistic p_Stat, int[][] p_Indice) {
StringBuilder featureBuilder = new StringBuilder();
svm_node[] features = this.getVector(p_Instance, p_Stat, p_Indice);
for (int i = 0; i < features.length; i++) {
featureBuilder.append(" ");
featureBuilder.append(features[i].index);
featureBuilder.append(":");
featureBuilder.append(features[i].value);
}
String featureOnly = featureBuilder.toString();
StringBuilder builder = new StringBuilder();
for (Integer tag : this.processTags(p_Stat, p_Instance.getTag())) {
builder.append(tag.toString());
builder.append(featureOnly);
builder.append("\n");
}
return builder.toString();
}
