public void printParentTopicAssignment(_ParentDoc2 d) {
String topicAssignmentFile =
"./data/results/0108_9/parentTopicAssignment/topicAssignment_" + d.getName() + "_.txt";
try {
PrintWriter pw = new PrintWriter(new File(topicAssignmentFile));
for (int i = 0; i < d.m_sentenceMap.size(); i++) {
_Stn stnObject = d.m_sentenceMap.get(i);
if (stnObject.m_stnLength != 0) {
int[] indexInCV = stnObject.m_words;
int[] positionInDoc = stnObject.m_wordPositionInDoc;
for (int j = 0; j < indexInCV.length; j++) {
int index = indexInCV[j];
int topic = d.m_topicAssignment[positionInDoc[j]];
String featureName = m_corpus.getFeature(index);
pw.print(featureName + ":" + topic + "\t");
}
}
pw.println();
}
pw.flush();
pw.close();
} catch (FileNotFoundException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
public double calParentLogLikelihood2(_ParentDoc2 pDoc) {
double likelihood = 0.0;
int tid = 0;
int wid = 0;
double term1 = 0.0;
double term2 = 0.0;
for (int n = 0; n < pDoc.getTotalDocLength(); n++) {
for (int k = 0; k < number_of_topics; k++) {
wid = pDoc.m_words[n];
// tid = pDoc.m_topicAssignment[n];
// normalize
term1 =
(m_parentWordTopicSstat[k][wid] + m_childWordTopicSstat[k][wid])
/ (double) (m_parentSstat[k] + m_childSstat[k]);
if (pDoc.getTotalDocLength() == 0) return 0;
term2 = pDoc.m_sstat[k] / (double) pDoc.getTotalDocLength();
// if (term2 == 0)
// System.out.println("term2 is zero");
// if (term1 == 0)
// System.out.println("term1 is zero");
if ((term1 == 0)) {
double eps = 1e-9;
term1 += eps;
}
if ((term2 == 0)) {
double eps = 1e-9;
term2 += eps;
}
likelihood += Math.log(term1) + Math.log(term2);
}
}
return likelihood;
}
// log space
public double calParentLogLikelihood(_ParentDoc2 pDoc) {
double term1 = 0.0;
double term2 = 0.0;
term1 = Utils.lgamma(number_of_topics * d_alpha) - number_of_topics * Utils.lgamma(d_alpha);
for (int k = 0; k < number_of_topics; k++) {
term2 += Utils.lgamma(pDoc.m_sstat[k] + d_alpha);
}
term2 -= Utils.lgamma((double) (number_of_topics * d_alpha + pDoc.getDocLength()));
return term1 + term2;
}
public void estStnThetaInParentDoc(_ParentDoc2 d) {
_SparseFeature[] fv = d.getSparse();
double[][] phi = new double[fv.length][number_of_topics];
HashMap<Integer, Integer> indexMap = new HashMap<Integer, Integer>();
// ///computeWordTopicProportionInDoc
for (int i = 0; i < fv.length; i++) {
int index = fv[i].getIndex();
indexMap.put(index, i);
}
for (int n = 0; n < d.m_words.length; n++) {
int index = d.m_words[n];
int topic = d.m_topicAssignment[n];
phi[indexMap.get(index)][topic]++;
}
for (int i = 0; i < fv.length; i++) {
Utils.L1Normalization(phi[i]);
}
for (int i = 0; i < d.m_sentenceMap.size(); i++) {
_Stn stnObject = d.m_sentenceMap.get(i);
stnObject.setTopicsVct(number_of_topics);
Arrays.fill(stnObject.m_topics, 0);
if (stnObject.m_stnLength != 0) {
int[] indexInCV = stnObject.m_words;
for (int j = 0; j < indexInCV.length; j++) {
int index = indexInCV[j];
for (int k = 0; k < number_of_topics; k++) {
stnObject.m_topics[k] += phi[indexMap.get(index)][k];
}
}
Utils.L1Normalization(stnObject.m_topics);
}
}
}