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

Example 16 with structures._SparseFeature

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

the class PRLogisticRegression method calcFuncGradient.

// This function is used to calculate the value and gradient with the new beta.
@Override
protected double calcFuncGradient(Collection<_Doc> trainSet) {
    double gValue = 0, fValue = 0;
    double Pij = 0, logPij = 0;
    // Add the L2 regularization.
    double L2 = 0, b;
    for (int i = 0; i < m_beta.length; i++) {
        b = m_beta[i];
        m_g[i] = 2 * m_lambda * b;
        L2 += b * b;
    }
    int Yi;
    _SparseFeature[] fv;
    int doc_index = 0;
    for (_Doc doc : trainSet) {
        fv = doc.getSparse();
        Yi = doc.getYLabel();
        // compute posterior regularized q(y=j|xi)\propto p(y=j|xi)exp(\lambda\phi(y=j, y)
        calcPosterior(fv, m_doc_pr[doc_index], m_cache);
        for (int j = 0; j < m_classNo; j++) {
            Pij = m_cache[j];
            logPij = Math.log(Pij);
            if (Yi == j) {
                gValue = Pij - 1;
                fValue += logPij;
            } else
                gValue = Pij;
            int offset = j * (m_featureSize + 1);
            m_g[offset] += gValue;
            // (Yij - Pij) * Xi
            for (_SparseFeature sf : fv) m_g[offset + sf.getIndex() + 1] += gValue * sf.getValue();
        }
        doc_index++;
    }
    return m_lambda * L2 - fValue;
}

Example 17 with structures._SparseFeature

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

the class LDAGibbs4AC_test method rankChild4ParentByLikelihood.

protected double rankChild4ParentByLikelihood(_ChildDoc cDoc, _ParentDoc pDoc) {
    int cDocLen = cDoc.getTotalDocLength();
    _SparseFeature[] fv = pDoc.getSparse();
    double docLogLikelihood = 0;
    for (_SparseFeature i : fv) {
        int wid = i.getIndex();
        double value = i.getValue();
        double wordLogLikelihood = 0;
        for (int k = 0; k < number_of_topics; k++) {
            double wordPerTopicLikelihood = (word_topic_sstat[k][wid] / m_sstat[k]) * ((cDoc.m_sstat[k] + d_alpha) / (d_alpha * number_of_topics + cDocLen));
            wordLogLikelihood += wordPerTopicLikelihood;
        }
        docLogLikelihood += value * Math.log(wordLogLikelihood);
    }
    return docLogLikelihood;
}

Example 18 with structures._SparseFeature

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

the class LDAGibbs4AC_test method rankChild4StnByLanguageModel.

protected HashMap<String, Double> rankChild4StnByLanguageModel(_Stn stnObj, _ParentDoc pDoc) {
    HashMap<String, Double> childLikelihoodMap = new HashMap<String, Double>();
    double smoothingMu = m_LM.m_smoothingMu;
    for (_ChildDoc cDoc : pDoc.m_childDocs) {
        int cDocLen = cDoc.getTotalDocLength();
        _SparseFeature[] fv = cDoc.getSparse();
        double stnLogLikelihood = 0;
        double alphaDoc = smoothingMu / (smoothingMu + cDocLen);
        _SparseFeature[] sv = stnObj.getFv();
        for (_SparseFeature svWord : sv) {
            double featureLikelihood = 0;
            int wid = svWord.getIndex();
            double stnVal = svWord.getValue();
            int featureIndex = Utils.indexOf(fv, wid);
            double docVal = 0;
            if (featureIndex != -1) {
                docVal = fv[featureIndex].getValue();
            }
            double smoothingProb = (1 - alphaDoc) * docVal / (cDocLen);
            smoothingProb += alphaDoc * m_LM.getReferenceProb(wid);
            featureLikelihood = Math.log(smoothingProb);
            stnLogLikelihood += stnVal * featureLikelihood;
        }
        childLikelihoodMap.put(cDoc.getName(), stnLogLikelihood);
    }
    return childLikelihoodMap;
}

Example 19 with structures._SparseFeature

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

the class LDAGibbs4AC_test method rankChild4StnByHybrid.

protected HashMap<String, Double> rankChild4StnByHybrid(_Stn stnObj, _ParentDoc pDoc) {
    HashMap<String, Double> childLikelihoodMap = new HashMap<String, Double>();
    double smoothingMu = m_LM.m_smoothingMu;
    for (_ChildDoc cDoc : pDoc.m_childDocs) {
        double cDocLen = cDoc.getTotalDocLength();
        _SparseFeature[] fv = cDoc.getSparse();
        double stnLogLikelihood = 0;
        double alphaDoc = smoothingMu / (smoothingMu + cDocLen);
        _SparseFeature[] sv = stnObj.getFv();
        for (_SparseFeature svWord : sv) {
            double featureLikelihood = 0;
            int wid = svWord.getIndex();
            double stnVal = svWord.getValue();
            int featureIndex = Utils.indexOf(fv, wid);
            double docVal = 0;
            if (featureIndex != -1) {
                docVal = fv[featureIndex].getValue();
            }
            double LMLikelihood = (1 - alphaDoc) * docVal / (cDocLen);
            LMLikelihood += alphaDoc * m_LM.getReferenceProb(wid);
            double TMLikelihood = 0;
            for (int k = 0; k < number_of_topics; k++) {
                // double likelihoodPerTopic =
                // topic_term_probabilty[k][wid];
                // System.out.println("likelihoodPerTopic1-----\t"+likelihoodPerTopic);
                // 
                // likelihoodPerTopic *= cDoc.m_topics[k];
                // System.out.println("likelihoodPerTopic2-----\t"+likelihoodPerTopic);
                TMLikelihood += (word_topic_sstat[k][wid] / m_sstat[k]) * (topicInDocProb(k, cDoc) / (d_alpha * number_of_topics + cDocLen));
            // TMLikelihood +=
            // topic_term_probabilty[k][wid]*cDoc.m_topics[k];
            // System.out.println("TMLikelihood\t"+TMLikelihood);
            }
            featureLikelihood = m_tau * LMLikelihood + (1 - m_tau) * TMLikelihood;
            // featureLikelihood = TMLikelihood;
            featureLikelihood = Math.log(featureLikelihood);
            stnLogLikelihood += stnVal * featureLikelihood;
        }
        childLikelihoodMap.put(cDoc.getName(), stnLogLikelihood);
    }
    return childLikelihoodMap;
}

Example 20 with structures._SparseFeature

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

the class corrLDA_Gibbs method calculate_log_likelihood4Parent.

protected double calculate_log_likelihood4Parent(_Doc d) {
    _ParentDoc pDoc = (_ParentDoc) d;
    double docLogLikelihood = 0;
    _SparseFeature[] fv = pDoc.getSparse();
    double docTopicSum = Utils.sumOfArray(pDoc.m_sstat);
    double alphaSum = d_alpha * number_of_topics;
    for (int j = 0; j < fv.length; j++) {
        int wid = fv[j].getIndex();
        double value = fv[j].getValue();
        double wordLogLikelihood = 0;
        for (int k = 0; k < number_of_topics; k++) {
            double wordPerTopicLikelihood = parentWordByTopicProb(k, wid) * parentTopicInDocProb(k, pDoc) / (alphaSum + docTopicSum);
            wordLogLikelihood += wordPerTopicLikelihood;
        }
        if (Math.abs(wordLogLikelihood) < 1e-10) {
            System.out.println("wordLogLikelihood\t" + wordLogLikelihood);
            wordLogLikelihood += 1e-10;
        }
        wordLogLikelihood = Math.log(wordLogLikelihood);
        docLogLikelihood += value * wordLogLikelihood;
    }
    return docLogLikelihood;
}