use of structures._ParentDoc in project IR_Base by Linda-sunshine.
the class DCMCorrLDA method updateAlphaC.
protected void updateAlphaC() {
double diff = 0;
int iteration = 0;
do {
diff = 0;
double totalAlphaDenominator = 0;
double[] totalAlphaNumerator = new double[number_of_topics];
Arrays.fill(totalAlphaNumerator, 0);
m_totalAlpha_c = Utils.sumOfArray(m_alpha_c);
double deltaAlpha = 0;
for (_Doc d : m_trainSet) {
if (d instanceof _ParentDoc) {
_ParentDoc4DCM pDoc = (_ParentDoc4DCM) d;
double pDocLen = pDoc.getTotalDocLength();
for (_ChildDoc cDoc : pDoc.m_childDocs) {
double muDp = cDoc.getMu() / pDocLen;
double t_totalAlpha_c = m_totalAlpha_c + cDoc.getMu();
double digAlpha = Utils.digamma(t_totalAlpha_c);
totalAlphaDenominator += Utils.digamma(cDoc.getTotalDocLength() + t_totalAlpha_c) - digAlpha;
for (int k = 0; k < number_of_topics; k++) totalAlphaNumerator[k] += Utils.digamma(m_alpha_c[k] + muDp * pDoc.m_sstat[k] + cDoc.m_sstat[k]) - Utils.digamma(m_alpha_c[k] + muDp * pDoc.m_sstat[k]);
}
}
}
for (int k = 0; k < number_of_topics; k++) {
deltaAlpha = totalAlphaNumerator[k] * 1.0 / totalAlphaDenominator;
double newAlpha = m_alpha_c[k] * deltaAlpha;
double t_diff = Math.abs(m_alpha_c[k] - newAlpha);
if (t_diff > diff)
diff = t_diff;
m_alpha_c[k] = newAlpha;
}
iteration++;
if (iteration > m_newtonIter)
break;
} while (diff > m_newtonConverge);
// System.out.println("iteration\t" + iteration);
m_totalAlpha_c = 0;
for (int k = 0; k < number_of_topics; k++) {
m_totalAlpha_c += m_alpha_c[k];
}
}
use of structures._ParentDoc in project IR_Base by Linda-sunshine.
the class DCMCorrLDA_multi_E method initialize_probability.
protected void initialize_probability(Collection<_Doc> collection) {
int cores = Runtime.getRuntime().availableProcessors();
m_threadpool = new Thread[cores];
m_workers = new DCMCorrLDA_worker[cores];
for (int i = 0; i < cores; i++) m_workers[i] = new DCMCorrLDA_worker(number_of_topics, vocabulary_size);
int workerID = 0;
for (_Doc d : collection) {
if (d instanceof _ParentDoc) {
m_workers[workerID % cores].addDoc(d);
_ParentDoc4DCM pDoc = (_ParentDoc4DCM) d;
for (_ChildDoc cDoc : pDoc.m_childDocs) {
m_workers[workerID % cores].addDoc(cDoc);
}
workerID++;
}
}
super.initialize_probability(collection);
}
use of structures._ParentDoc in project IR_Base by Linda-sunshine.
the class DCMCorrLDA_multi_E_test method debugOutput.
protected void debugOutput(int topK, String filePrefix) {
File parentTopicFolder = new File(filePrefix + "parentTopicAssignment");
File childTopicFolder = new File(filePrefix + "childTopicAssignment");
if (!parentTopicFolder.exists()) {
System.out.println("creating directory" + parentTopicFolder);
parentTopicFolder.mkdir();
}
if (!childTopicFolder.exists()) {
System.out.println("creating directory" + childTopicFolder);
childTopicFolder.mkdir();
}
File parentWordTopicDistributionFolder = new File(filePrefix + "wordTopicDistribution");
if (!parentWordTopicDistributionFolder.exists()) {
System.out.println("creating word topic distribution folder\t" + parentWordTopicDistributionFolder);
parentWordTopicDistributionFolder.mkdir();
}
for (_Doc d : m_trainSet) {
if (d instanceof _ParentDoc) {
// printParentTopicAssignment(d, parentTopicFolder);
printWordTopicDistribution(d, parentWordTopicDistributionFolder, topK);
} else {
printChildTopicAssignment(d, childTopicFolder);
}
}
String parentParameterFile = filePrefix + "parentParameter.txt";
String childParameterFile = filePrefix + "childParameter.txt";
printParameter(parentParameterFile, childParameterFile, m_trainSet);
printTopKChild4Stn(filePrefix, topK);
}
use of structures._ParentDoc in project IR_Base by Linda-sunshine.
the class DCMCorrLDA_test method printParameter.
protected void printParameter(String parentParameterFile, String childParameterFile, ArrayList<_Doc> docList) {
System.out.println("printing parameter");
try {
System.out.println(parentParameterFile);
System.out.println(childParameterFile);
PrintWriter parentParaOut = new PrintWriter(new File(parentParameterFile));
PrintWriter childParaOut = new PrintWriter(new File(childParameterFile));
for (_Doc d : docList) {
if (d instanceof _ParentDoc) {
parentParaOut.print(d.getName() + "\t");
parentParaOut.print("topicProportion\t");
for (int k = 0; k < number_of_topics; k++) {
parentParaOut.print(d.m_topics[k] + "\t");
}
parentParaOut.println();
for (_ChildDoc cDoc : ((_ParentDoc) d).m_childDocs) {
childParaOut.print(cDoc.getName() + "\t");
childParaOut.print("topicProportion\t");
for (int k = 0; k < number_of_topics; k++) {
childParaOut.print(cDoc.m_topics[k] + "\t");
}
childParaOut.println();
}
}
}
parentParaOut.flush();
parentParaOut.close();
childParaOut.flush();
childParaOut.close();
} catch (Exception e) {
e.printStackTrace();
}
}
use of structures._ParentDoc in project IR_Base by Linda-sunshine.
the class ParentChildAnalyzer method loadParentDoc.
public void loadParentDoc(String fileName) {
if (fileName == null || fileName.isEmpty())
return;
JSONObject json = LoadJSON(fileName);
String title = Utils.getJSONValue(json, "title");
String content = Utils.getJSONValue(json, "content");
String name = Utils.getJSONValue(json, "name");
String[] sentences = null;
// _ParentDoc d = new _ParentDoc(m_corpus.getSize(), name, title,
// content, 0);
_ParentDoc d = new _ParentDoc4DCM(m_corpus.getSize(), name, title, content, 0);
try {
JSONArray sentenceArray = json.getJSONArray("sentences");
sentences = new String[sentenceArray.length()];
// shall we add title into this sentence array
for (int i = 0; i < sentenceArray.length(); i++) sentences[i] = Utils.getJSONValue(sentenceArray.getJSONObject(i), "sentence");
if (AnalyzeDocByStn(d, sentences))
parentHashMap.put(name, d);
} catch (JSONException e) {
e.printStackTrace();
}
}
Aggregations