Examples with MultiLabel edu.neu.ccs.pyramid.dataset.MultiLabel used on opensource projects

Example 96 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class CMLCRFTest method test2.

public static void test2() throws Exception {
    System.out.println(config);
    MultiLabelClfDataSet trainSet = TRECFormat.loadMultiLabelClfDataSet(config.getString("input.trainData"), DataSetType.ML_CLF_DENSE, true);
    MultiLabelClfDataSet testSet = TRECFormat.loadMultiLabelClfDataSet(config.getString("input.testData"), DataSetType.ML_CLF_DENSE, true);
    double gaussianVariance = config.getDouble("gaussianVariance");
    // loading or save model infos.
    String output = config.getString("output");
    String modelName = config.getString("modelName");
    CMLCRF cmlcrf;
    MultiLabel[] predTrain;
    MultiLabel[] predTest;
    if (config.getBoolean("train.warmStart")) {
        cmlcrf = CMLCRF.deserialize(new File(output, modelName));
        System.out.println("loading model:");
        System.out.println(cmlcrf);
    } else {
        cmlcrf = new CMLCRF(trainSet);
        CRFLoss crfLoss = new CRFLoss(cmlcrf, trainSet, gaussianVariance);
        if (config.getBoolean("isLBFGS")) {
            LBFGS optimizer = new LBFGS(crfLoss);
            optimizer.getTerminator().setAbsoluteEpsilon(0.1);
            for (int i = 0; i < config.getInt("numRounds"); i++) {
                optimizer.iterate();
                predTrain = cmlcrf.predict(trainSet);
                predTest = cmlcrf.predict(testSet);
                System.out.print("iter: " + String.format("%04d", i));
                System.out.print("\tTrain acc: " + String.format("%.4f", Accuracy.accuracy(trainSet.getMultiLabels(), predTrain)));
                System.out.print("\tTrain overlap " + String.format("%.4f", Overlap.overlap(trainSet.getMultiLabels(), predTrain)));
                System.out.print("\tTest acc: " + String.format("%.4f", Accuracy.accuracy(testSet.getMultiLabels(), predTest)));
                System.out.println("\tTest overlap " + String.format("%.4f", Overlap.overlap(testSet.getMultiLabels(), predTest)));
            }
        } else {
            GradientDescent optimizer = new GradientDescent(crfLoss);
            for (int i = 0; i < config.getInt("numRounds"); i++) {
                optimizer.iterate();
                predTrain = cmlcrf.predict(trainSet);
                predTest = cmlcrf.predict(testSet);
                System.out.print("iter: " + String.format("%04d", i));
                System.out.print("\tTrain acc: " + String.format("%.4f", Accuracy.accuracy(trainSet.getMultiLabels(), predTrain)));
                System.out.print("\tTrain overlap " + String.format("%.4f", Overlap.overlap(trainSet.getMultiLabels(), predTrain)));
                System.out.print("\tTest acc: " + String.format("%.4f", Accuracy.accuracy(testSet.getMultiLabels(), predTest)));
                System.out.println("\tTest overlap " + String.format("%.4f", Overlap.overlap(testSet.getMultiLabels(), predTest)));
            }
        }
    }
    System.out.println();
    System.out.println();
    System.out.println("--------------------------------Results-----------------------------\n");
    predTrain = cmlcrf.predict(trainSet);
    predTest = cmlcrf.predict(testSet);
    System.out.print("Train acc: " + String.format("%.4f", Accuracy.accuracy(trainSet.getMultiLabels(), predTrain)));
    System.out.print("\tTrain overlap " + String.format("%.4f", Overlap.overlap(trainSet.getMultiLabels(), predTrain)));
    System.out.print("\tTest acc: " + String.format("%.4f", Accuracy.accuracy(testSet.getMultiLabels(), predTest)));
    System.out.println("\tTest overlap " + String.format("%.4f", Overlap.overlap(testSet.getMultiLabels(), predTest)));
    if (config.getBoolean("saveModel")) {
        (new File(output)).mkdirs();
        File serializeModel = new File(output, modelName);
        cmlcrf.serialize(serializeModel);
    }
}

Example 97 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class CMLCRFTest method test7.

private static void test7() throws Exception {
    System.out.println(config);
    MultiLabelClfDataSet trainSet = TRECFormat.loadMultiLabelClfDataSet(config.getString("input.trainData"), DataSetType.ML_CLF_SEQ_SPARSE, true);
    MultiLabelClfDataSet testSet = TRECFormat.loadMultiLabelClfDataSet(config.getString("input.testData"), DataSetType.ML_CLF_SEQ_SPARSE, true);
    // loading or save model infos.
    String output = config.getString("output");
    String modelName = config.getString("modelName");
    CMLCRF cmlcrf = null;
    if (config.getString("train.warmStart").equals("true")) {
        cmlcrf = CMLCRF.deserialize(new File(output, modelName));
        System.out.println("loading model:");
        System.out.println(cmlcrf);
    } else if (config.getString("train.warmStart").equals("auto")) {
        cmlcrf = CMLCRF.deserialize(new File(output, modelName));
        System.out.println("retrain model:");
        CMLCRFElasticNet cmlcrfElasticNet = new CMLCRFElasticNet(cmlcrf, trainSet, config.getDouble("l1Ratio"), config.getDouble("regularization"));
        train(cmlcrfElasticNet, cmlcrf, trainSet, testSet, config);
    } else if (config.getString("train.warmStart").equals("false")) {
        cmlcrf = new CMLCRF(trainSet);
        cmlcrf.setConsiderPair(config.getBoolean("considerLabelPair"));
        CMLCRFElasticNet cmlcrfElasticNet = new CMLCRFElasticNet(cmlcrf, trainSet, config.getDouble("l1Ratio"), config.getDouble("regularization"));
        train(cmlcrfElasticNet, cmlcrf, trainSet, testSet, config);
    }
    System.out.println();
    System.out.println();
    System.out.println("--------------------------------Results-----------------------------\n");
    MLMeasures measures = new MLMeasures(cmlcrf, trainSet);
    System.out.println("========== Train ==========\n");
    System.out.println(measures);
    System.out.println("========== Test ==========\n");
    long startTimePred = System.nanoTime();
    MultiLabel[] preds = cmlcrf.predict(testSet);
    long stopTimePred = System.nanoTime();
    long predTime = stopTimePred - startTimePred;
    System.out.println("\nprediction time: " + TimeUnit.NANOSECONDS.toSeconds(predTime) + " sec.");
    System.out.println(new MLMeasures(cmlcrf, testSet));
    System.out.println("\n\n");
    InstanceF1Predictor pluginF1 = new InstanceF1Predictor(cmlcrf);
    System.out.println("Plugin F1");
    System.out.println(new MLMeasures(pluginF1, testSet));
    if (config.getBoolean("saveModel")) {
        (new File(output)).mkdirs();
        File serializeModel = new File(output, modelName);
        cmlcrf.serialize(serializeModel);
    }
}

Example 98 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class CBMInspectorTest method test1.

private static void test1() throws Exception {
    MultiLabelClfDataSet testSet = TRECFormat.loadMultiLabelClfDataSet(new File(DATASETS, "meka_imdb/1/data_sets/test"), DataSetType.ML_CLF_SPARSE, true);
    CBM CBM = (CBM) Serialization.deserialize(new File(TMP, "model"));
    System.out.println(Accuracy.accuracy(CBM, testSet));
    for (int i = 0; i < testSet.getNumDataPoints(); i++) {
        MultiLabel trueLabel = testSet.getMultiLabels()[i];
        MultiLabel pred = CBM.predict(testSet.getRow(i));
        MultiLabel expectation = CBM.predictByMarginals(testSet.getRow(i));
        if (pred.equals(trueLabel) && !pred.equals(expectation) && expectation.getMatchedLabels().size() > 0) {
            System.out.println("==============================");
            System.out.println("data point " + i);
            System.out.println("prediction = " + pred);
            System.out.println("expectation = " + expectation);
            CBMInspector.covariance(CBM, testSet.getRow(i), testSet.getLabelTranslator());
        }
    }
}

Example 99 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class CMLCRFTest method test5.

private static void test5() throws Exception {
    MultiLabelClfDataSet dataSet = TRECFormat.loadMultiLabelClfDataSet(new File(DATASETS, "ohsumed/3/train.trec"), DataSetType.ML_CLF_SPARSE, true);
    MultiLabelClfDataSet testSet = TRECFormat.loadMultiLabelClfDataSet(new File(DATASETS, "ohsumed/3/test.trec"), DataSetType.ML_CLF_SPARSE, true);
    CMLCRF cmlcrf = new CMLCRF(dataSet);
    CRFLoss crfLoss = new CRFLoss(cmlcrf, dataSet, 1);
    cmlcrf.setConsiderPair(false);
    MultiLabel[] predTrain;
    MultiLabel[] predTest;
    LBFGS optimizer = new LBFGS(crfLoss);
    for (int i = 0; i < 5; i++) {
        //            System.out.print("Obj: " + optimizer.getTerminator().getLastValue());
        System.out.println("iter: " + i);
        optimizer.iterate();
        System.out.println(crfLoss.getValue());
        predTrain = cmlcrf.predict(dataSet);
        predTest = cmlcrf.predict(testSet);
        System.out.print("\tTrain acc: " + Accuracy.accuracy(dataSet.getMultiLabels(), predTrain));
        System.out.print("\tTrain overlap " + Overlap.overlap(dataSet.getMultiLabels(), predTrain));
        System.out.print("\tTest acc: " + Accuracy.accuracy(testSet.getMultiLabels(), predTest));
        System.out.println("\tTest overlap " + Overlap.overlap(testSet.getMultiLabels(), predTest));
    //            System.out.println("crf = "+cmlcrf.getWeights());
    //            System.out.println(Arrays.toString(predTrain));
    }
    CRFLoss crfLoss2 = new CRFLoss(cmlcrf, dataSet, 1);
    cmlcrf.setConsiderPair(true);
    LBFGS optimizer2 = new LBFGS(crfLoss2);
    for (int i = 0; i < 50; i++) {
        System.out.println("consider pairs");
        //            System.out.print("Obj: " + optimizer.getTerminator().getLastValue());
        System.out.println("iter: " + i);
        optimizer2.iterate();
        System.out.println(crfLoss2.getValue());
        predTrain = cmlcrf.predict(dataSet);
        predTest = cmlcrf.predict(testSet);
        System.out.print("\tTrain acc: " + Accuracy.accuracy(dataSet.getMultiLabels(), predTrain));
        System.out.print("\tTrain overlap " + Overlap.overlap(dataSet.getMultiLabels(), predTrain));
        System.out.print("\tTest acc: " + Accuracy.accuracy(testSet.getMultiLabels(), predTest));
        System.out.println("\tTest overlap " + Overlap.overlap(testSet.getMultiLabels(), predTest));
    //            System.out.println("crf = "+cmlcrf.getWeights());
    //            System.out.println(Arrays.toString(predTrain));
    }
}

Example 100 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class CMLCRFTest method test3.

private static void test3() throws Exception {
    MultiLabelClfDataSet dataSet = TRECFormat.loadMultiLabelClfDataSet(new File(DATASETS, "/imdb/3/train.trec"), DataSetType.ML_CLF_SPARSE, true);
    MultiLabelClfDataSet testSet = TRECFormat.loadMultiLabelClfDataSet(new File(DATASETS, "/imdb/3/test.trec"), DataSetType.ML_CLF_SPARSE, true);
    CMLCRF cmlcrf = new CMLCRF(dataSet);
    CRFLoss crfLoss = new CRFLoss(cmlcrf, dataSet, 1);
    MultiLabel[] predTrain;
    MultiLabel[] predTest;
    LBFGS optimizer = new LBFGS(crfLoss);
    for (int i = 0; i < 50; i++) {
        //            System.out.print("Obj: " + optimizer.getTerminator().getLastValue());
        System.out.println("iter: " + i);
        optimizer.iterate();
        System.out.println(crfLoss.getValue());
        predTrain = cmlcrf.predict(dataSet);
        predTest = cmlcrf.predict(testSet);
        System.out.print("\tTrain acc: " + Accuracy.accuracy(dataSet.getMultiLabels(), predTrain));
        System.out.print("\tTrain overlap " + Overlap.overlap(dataSet.getMultiLabels(), predTrain));
        System.out.print("\tTest acc: " + Accuracy.accuracy(testSet.getMultiLabels(), predTest));
        System.out.println("\tTest overlap " + Overlap.overlap(testSet.getMultiLabels(), predTest));
    //            System.out.println("crf = "+cmlcrf.getWeights());
    //            System.out.println(Arrays.toString(predTrain));
    }
}