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

Example 1 with structures._PerformanceStat

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

the class ModelAdaptation method test.

@Override
public double test() {
    int numberOfCores = Runtime.getRuntime().availableProcessors();
    ArrayList<Thread> threads = new ArrayList<Thread>();
    for (int k = 0; k < numberOfCores; ++k) {
        threads.add((new Thread() {

            int core, numOfCores;

            @Override
            public void run() {
                _AdaptStruct user;
                _PerformanceStat userPerfStat;
                try {
                    for (int i = 0; i + core < m_userList.size(); i += numOfCores) {
                        user = m_userList.get(i + core);
                        if (// no testing data
                        (m_testmode == TestMode.TM_batch && user.getTestSize() < 1) || // no adaptation data
                        (m_testmode == TestMode.TM_online && user.getAdaptationSize() < 1) || // no testing and adaptation data
                        (m_testmode == TestMode.TM_hybrid && user.getAdaptationSize() < 1) && user.getTestSize() < 1)
                            continue;
                        userPerfStat = user.getPerfStat();
                        if (m_testmode == TestMode.TM_batch || m_testmode == TestMode.TM_hybrid) {
                            // record prediction results
                            for (_Review r : user.getReviews()) {
                                if (r.getType() != rType.TEST)
                                    continue;
                                int trueL = r.getYLabel();
                                // evoke user's own model
                                int predL = user.predict(r);
                                r.setPredictLabel(predL);
                                userPerfStat.addOnePredResult(predL, trueL);
                            }
                        }
                        userPerfStat.calculatePRF();
                    }
                } catch (Exception ex) {
                    ex.printStackTrace();
                }
            }

            private Thread initialize(int core, int numOfCores) {
                this.core = core;
                this.numOfCores = numOfCores;
                return this;
            }
        }).initialize(k, numberOfCores));
        threads.get(k).start();
    }
    for (int k = 0; k < numberOfCores; ++k) {
        try {
            threads.get(k).join();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
    }
    int count = 0;
    ArrayList<ArrayList<Double>> macroF1 = new ArrayList<ArrayList<Double>>();
    // init macroF1
    for (int i = 0; i < m_classNo; i++) macroF1.add(new ArrayList<Double>());
    _PerformanceStat userPerfStat;
    m_microStat.clear();
    for (_AdaptStruct user : m_userList) {
        if (// no testing data
        (m_testmode == TestMode.TM_batch && user.getTestSize() < 1) || // no adaptation data
        (m_testmode == TestMode.TM_online && user.getAdaptationSize() < 1) || // no testing and adaptation data
        (m_testmode == TestMode.TM_hybrid && user.getAdaptationSize() < 1) && user.getTestSize() < 1)
            continue;
        userPerfStat = user.getPerfStat();
        for (int i = 0; i < m_classNo; i++) {
            if (userPerfStat.getTrueClassNo(i) > 0)
                macroF1.get(i).add(userPerfStat.getF1(i));
        }
        m_microStat.accumulateConfusionMat(userPerfStat);
        count++;
    }
    System.out.print("neg users: " + macroF1.get(0).size());
    System.out.print("\tpos users: " + macroF1.get(1).size() + "\n");
    System.out.println(toString());
    calcMicroPerfStat();
    // macro average and standard deviation.
    System.out.println("\nMacro F1:");
    for (int i = 0; i < m_classNo; i++) {
        double[] avgStd = calcAvgStd(macroF1.get(i));
        m_perf[i] = avgStd[0];
        System.out.format("Class %d: %.4f+%.4f\t", i, avgStd[0], avgStd[1]);
    }
    // printPerformance();
    return 0;
}

Example 2 with structures._PerformanceStat

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

the class asyncRegLR method train.

// this is online training in each individual user
@Override
public double train() {
    double gNorm, gNormOld = Double.MAX_VALUE;
    ;
    int predL, trueL;
    _Review doc;
    _PerformanceStat perfStat;
    initLBFGS();
    init();
    for (_AdaptStruct user : m_userList) {
        while (user.hasNextAdaptationIns()) {
            // test the latest model before model adaptation
            if (m_testmode != TestMode.TM_batch && (doc = user.getLatestTestIns()) != null) {
                perfStat = user.getPerfStat();
                predL = predict(doc, user);
                trueL = doc.getYLabel();
                perfStat.addOnePredResult(predL, trueL);
            }
            // in batch mode we will not accumulate the performance during adaptation
            // prepare to adapt: initialize gradient
            Arrays.fill(m_g, 0);
            calculateGradients(user);
            gNorm = gradientTest();
            if (m_displayLv == 1) {
                if (gNorm < gNormOld)
                    System.out.print("o");
                else
                    System.out.print("x");
            }
            // gradient descent
            gradientDescent(user, m_initStepSize, m_g);
            gNormOld = gNorm;
        }
        if (m_displayLv > 0)
            System.out.println();
    }
    setPersonalizedModel();
    // we do not evaluate function value
    return 0;
}

Example 3 with structures._PerformanceStat

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

the class asyncCoLinAdapt method train.

// this is online training in each individual user
@Override
public double train() {
    double gNorm, gNormOld = Double.MAX_VALUE;
    int updateCount = 0;
    _CoLinAdaptStruct user;
    int predL, trueL;
    _Review doc;
    _PerformanceStat perfStat;
    initLBFGS();
    init();
    for (int t = 0; t < m_userOrder.length; t++) {
        user = (_CoLinAdaptStruct) m_userList.get(m_userOrder[t]);
        if (user.hasNextAdaptationIns()) {
            // test the latest model
            if (m_testmode != TestMode.TM_batch && (doc = user.getLatestTestIns()) != null) {
                perfStat = user.getPerfStat();
                predL = predict(doc, user);
                trueL = doc.getYLabel();
                perfStat.addOnePredResult(predL, trueL);
            }
            // in batch mode we will not accumulate the performance during adaptation
            // prepare to adapt: initialize gradient
            Arrays.fill(m_g, 0);
            calculateGradients(user);
            gNorm = gradientTest(user);
            if (m_displayLv == 1) {
                if (gNorm < gNormOld)
                    System.out.print("o");
                else
                    System.out.print("x");
            }
            // gradient descent
            gradientDescent(user, m_initStepSize, 1.0);
            // gradientDescent(user, asyncLinAdapt.getStepSize(initStepSize, user));
            gNormOld = gNorm;
            if (m_displayLv > 0 && ++updateCount % 100 == 0)
                System.out.println();
        }
    }
    if (m_displayLv > 0)
        System.out.println();
    setPersonalizedModel();
    // we do not evaluate function value
    return 0;
}

Example 4 with structures._PerformanceStat

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

the class asyncLinAdapt method train.

// this is online training in each individual user
@Override
public double train() {
    double gNorm, gNormOld = Double.MAX_VALUE;
    ;
    int predL, trueL;
    _Review doc;
    _PerformanceStat perfStat;
    _LinAdaptStruct user;
    initLBFGS();
    init();
    for (int i = 0; i < m_userList.size(); i++) {
        user = (_LinAdaptStruct) m_userList.get(i);
        while (user.hasNextAdaptationIns()) {
            // test the latest model before model adaptation
            if (m_testmode != TestMode.TM_batch && (doc = user.getLatestTestIns()) != null) {
                perfStat = user.getPerfStat();
                predL = predict(doc, user);
                trueL = doc.getYLabel();
                perfStat.addOnePredResult(predL, trueL);
            }
            // in batch mode we will not accumulate the performance during adaptation
            // prepare to adapt: initialize gradient
            Arrays.fill(m_g, 0);
            calculateGradients(user);
            gNorm = gradientTest();
            if (m_displayLv == 1) {
                if (gNorm < gNormOld)
                    System.out.print("o");
                else
                    System.out.print("x");
            }
            // gradient descent
            asyncRegLR.gradientDescent(user, m_initStepSize, m_g);
            gNormOld = gNorm;
        }
        if (m_displayLv > 0)
            System.out.println();
    }
    setPersonalizedModel();
    // we do not evaluate function value
    return 0;
}

Example 5 with structures._PerformanceStat

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

the class ModelAdaptation method printPerformance.

public void printPerformance() {
    _PerformanceStat perf;
    for (_AdaptStruct user : m_userList) {
        perf = user.getPerfStat();
        System.out.print(String.format("pos:%d\tneg:%d\tposF1:%.4f\tnegF1:%.4f\n", perf.getTrueClassNo(1), perf.getTrueClassNo(0), perf.getF1(1), perf.getF1(0)));
    }
}