Examples with structures._ParentDoc4DCM structures._ParentDoc4DCM used on opensource projects

Example 26 with structures._ParentDoc4DCM

use of structures._ParentDoc4DCM in project IR_Base by Linda-sunshine.

the class DCMCorrLDA_test method rankChild4StnByLikelihood.

protected HashMap<String, Double> rankChild4StnByLikelihood(_Stn stnObj, _ParentDoc4DCM pDoc) {
    HashMap<String, Double> likelihoodMap = new HashMap<String, Double>();
    for (_ChildDoc cDoc : pDoc.m_childDocs) {
        double stnLogLikelihood = 0;
        for (_Word w : stnObj.getWords()) {
            double wordLikelihood = 0;
            int wid = w.getIndex();
            for (int k = 0; k < number_of_topics; k++) {
                wordLikelihood += childTopicInDocProb(k, cDoc, pDoc) * childWordByTopicProb(k, wid, pDoc);
            }
            stnLogLikelihood += wordLikelihood;
        }
        likelihoodMap.put(cDoc.getName(), stnLogLikelihood);
    }
    return likelihoodMap;
}

Example 27 with structures._ParentDoc4DCM

use of structures._ParentDoc4DCM in project IR_Base by Linda-sunshine.

the class DCMCorrLDA_test method printWordTopicDistribution.

protected void printWordTopicDistribution(_Doc d, File wordTopicDistributionFolder, int k) {
    _ParentDoc4DCM pDoc = (_ParentDoc4DCM) d;
    String wordTopicDistributionFile = pDoc.getName() + ".txt";
    try {
        PrintWriter pw = new PrintWriter(new File(wordTopicDistributionFolder, wordTopicDistributionFile));
        for (int i = 0; i < number_of_topics; i++) {
            MyPriorityQueue<_RankItem> fVector = new MyPriorityQueue<_RankItem>(k);
            for (int v = 0; v < vocabulary_size; v++) {
                String featureName = m_corpus.getFeature(v);
                double wordProb = pDoc.m_wordTopic_prob[i][v];
                _RankItem ri = new _RankItem(featureName, wordProb);
                fVector.add(ri);
            }
            pw.format("Topic %d(%.5f):\t", i, pDoc.m_topics[i]);
            for (_RankItem it : fVector) pw.format("%s(%.5f)\t", it.m_name, m_logSpace ? Math.exp(it.m_value) : it.m_value);
            pw.write("\n");
        }
        pw.flush();
        pw.close();
    } catch (FileNotFoundException e) {
        e.printStackTrace();
    }
}

Example 28 with structures._ParentDoc4DCM

use of structures._ParentDoc4DCM in project IR_Base by Linda-sunshine.

the class DCMLDA4AC method calculate_M_step.

public void calculate_M_step(int iter, File weightFolder) {
    for (_Doc d : m_trainSet) {
        if (d instanceof _ParentDoc4DCM)
            collectParentStats((_ParentDoc4DCM) d);
        else
            collectChildStats((_ChildDoc) d);
    }
    for (int k = 0; k < number_of_topics; k++) for (int v = 0; v < vocabulary_size; v++) m_topic_word_prob[k][v] += word_topic_sstat[k][v] + m_beta[k][v];
    File weightIterFolder = new File(weightFolder, "_" + iter);
    if (!weightIterFolder.exists()) {
        weightIterFolder.mkdir();
    }
    updateParameter(iter, weightIterFolder);
}

Example 29 with structures._ParentDoc4DCM

use of structures._ParentDoc4DCM in project IR_Base by Linda-sunshine.

the class DCMLDA4AC method initialize_probability.

protected void initialize_probability(Collection<_Doc> collection) {
    m_alpha = new double[number_of_topics];
    m_beta = new double[number_of_topics][vocabulary_size];
    m_totalAlpha = 0;
    m_totalBeta = new double[number_of_topics];
    m_topic_word_prob = new double[number_of_topics][vocabulary_size];
    for (_Doc d : collection) {
        if (d instanceof _ParentDoc4DCM) {
            _ParentDoc4DCM pDoc = (_ParentDoc4DCM) d;
            pDoc.setTopics4Gibbs(number_of_topics, 0, vocabulary_size);
            for (_Word w : pDoc.getWords()) {
                int wid = w.getIndex();
                int tid = w.getTopic();
                word_topic_sstat[tid][wid]++;
            }
            for (_ChildDoc cDoc : pDoc.m_childDocs) {
                cDoc.setTopics4Gibbs_LDA(number_of_topics, 0);
                for (_Word w : cDoc.getWords()) {
                    int wid = w.getIndex();
                    int tid = w.getTopic();
                    pDoc.m_wordTopic_stat[tid][wid]++;
                    pDoc.m_topic_stat[tid]++;
                    word_topic_sstat[tid][wid]++;
                }
            }
        }
    }
    initialAlphaBeta();
    imposePrior();
}

Example 30 with structures._ParentDoc4DCM

use of structures._ParentDoc4DCM in project IR_Base by Linda-sunshine.

the class DCMLDA4AC method calculate_log_likelihood.

protected double calculate_log_likelihood(_ParentDoc4DCM d) {
    double docLogLikelihood = 0.0;
    double parentDocLength = d.getTotalDocLength();
    for (int k = 0; k < number_of_topics; k++) {
        double term = Utils.lgamma(d.m_sstat[k] + m_alpha[k]);
        docLogLikelihood += term;
        term = Utils.lgamma(m_alpha[k]);
        docLogLikelihood -= term;
    }
    docLogLikelihood += Utils.lgamma(m_totalAlpha);
    docLogLikelihood -= Utils.lgamma(parentDocLength + m_totalAlpha);
    for (int k = 0; k < number_of_topics; k++) {
        for (int v = 0; v < vocabulary_size; v++) {
            double term = Utils.lgamma(d.m_wordTopic_stat[k][v] + m_beta[k][v]);
            docLogLikelihood += term;
            term = Utils.lgamma(m_beta[k][v]);
            docLogLikelihood -= term;
        }
        docLogLikelihood += Utils.lgamma(m_totalBeta[k]);
        docLogLikelihood -= Utils.lgamma(d.m_topic_stat[k] + m_totalBeta[k]);
    }
    for (_ChildDoc cDoc : d.m_childDocs) {
        int cDocLength = cDoc.getTotalDocLength();
        for (int k = 0; k < number_of_topics; k++) {
            double term = Utils.lgamma(cDoc.m_sstat[k] + m_alpha[k]);
            docLogLikelihood += term;
            term = Utils.lgamma(m_alpha[k]);
            docLogLikelihood -= term;
        }
        docLogLikelihood += Utils.lgamma(m_totalAlpha);
        docLogLikelihood -= Utils.lgamma(cDocLength + m_totalAlpha);
    }
    return docLogLikelihood;
}