Examples with RegTreeFactory edu.neu.ccs.pyramid.regression.regression_tree.RegTreeFactory used on opensource projects

Example 11 with RegTreeFactory

use of edu.neu.ccs.pyramid.regression.regression_tree.RegTreeFactory in project pyramid by cheng-li.

the class CBMOptimizer method updateBinaryBoosting.

private void updateBinaryBoosting(int componentIndex, int labelIndex) {
    int numIterations = numIterationsBinary;
    double shrinkage = shrinkageBinary;
    LKBoost boost = (LKBoost) this.cbm.binaryClassifiers[componentIndex][labelIndex];
    RegTreeConfig regTreeConfig = new RegTreeConfig().setMaxNumLeaves(numLeavesBinary);
    RegTreeFactory regTreeFactory = new RegTreeFactory(regTreeConfig);
    regTreeFactory.setLeafOutputCalculator(new LKBOutputCalculator(2));
    LKBoostOptimizer optimizer = new LKBoostOptimizer(boost, dataSet, regTreeFactory, gammasT[componentIndex], targetsDistributions[labelIndex]);
    optimizer.setShrinkage(shrinkage);
    optimizer.initialize();
    optimizer.iterate(numIterations);
}

Example 12 with RegTreeFactory

use of edu.neu.ccs.pyramid.regression.regression_tree.RegTreeFactory in project pyramid by cheng-li.

the class GBCBMOptimizer method updateBinaryClassifier.

@Override
protected void updateBinaryClassifier(int component, int label, MultiLabelClfDataSet activeDataset, double[] activeGammas) {
    StopWatch stopWatch = new StopWatch();
    stopWatch.start();
    if (cbm.binaryClassifiers[component][label] == null || cbm.binaryClassifiers[component][label] instanceof PriorProbClassifier) {
        cbm.binaryClassifiers[component][label] = new LKBoost(2);
    }
    int[] binaryLabels = DataSetUtil.toBinaryLabels(activeDataset.getMultiLabels(), label);
    double[][] targetsDistributions = DataSetUtil.labelsToDistributions(binaryLabels, 2);
    LKBoost boost = (LKBoost) this.cbm.binaryClassifiers[component][label];
    RegTreeConfig regTreeConfig = new RegTreeConfig().setMaxNumLeaves(numLeaves);
    RegTreeFactory regTreeFactory = new RegTreeFactory(regTreeConfig);
    regTreeFactory.setLeafOutputCalculator(new LKBOutputCalculator(2));
    LKBoostOptimizer optimizer = new LKBoostOptimizer(boost, activeDataset, regTreeFactory, activeGammas, targetsDistributions);
    optimizer.setShrinkage(shrinkage);
    optimizer.initialize();
    optimizer.iterate(binaryUpdatesPerIter);
    if (logger.isDebugEnabled()) {
        logger.debug("time spent on updating component " + component + " label " + label + " = " + stopWatch);
    }
}

Example 13 with RegTreeFactory

use of edu.neu.ccs.pyramid.regression.regression_tree.RegTreeFactory in project pyramid by cheng-li.

the class LSBoostTest method test1.

private static void test1() throws Exception {
    RegDataSet trainSet = TRECFormat.loadRegDataSet(new File(DATASETS, "abalone/folds/fold_1/train.trec"), DataSetType.REG_DENSE, true);
    RegDataSet testSet = TRECFormat.loadRegDataSet(new File(DATASETS, "abalone/folds/fold_1/test.trec"), DataSetType.REG_DENSE, true);
    LSBoost lsBoost = new LSBoost();
    RegTreeConfig regTreeConfig = new RegTreeConfig().setMaxNumLeaves(3);
    RegTreeFactory regTreeFactory = new RegTreeFactory(regTreeConfig);
    LSBoostOptimizer optimizer = new LSBoostOptimizer(lsBoost, trainSet, regTreeFactory);
    optimizer.setShrinkage(0.1);
    optimizer.initialize();
    for (int i = 0; i < 100; i++) {
        System.out.println("iteration " + i);
        System.out.println("train RMSE = " + RMSE.rmse(lsBoost, trainSet));
        System.out.println("test RMSE = " + RMSE.rmse(lsBoost, testSet));
        optimizer.iterate();
    }
}

Example 14 with RegTreeFactory

use of edu.neu.ccs.pyramid.regression.regression_tree.RegTreeFactory in project pyramid by cheng-li.

the class GBRegressor method train.

private static void train(Config config) throws Exception {
    String sparsity = config.getString("input.matrixType");
    DataSetType dataSetType = null;
    switch(sparsity) {
        case "dense":
            dataSetType = DataSetType.REG_DENSE;
            break;
        case "sparse":
            dataSetType = DataSetType.REG_SPARSE;
            break;
        default:
            throw new IllegalArgumentException("input.matrixType should be dense or sparse");
    }
    RegDataSet trainSet = TRECFormat.loadRegDataSet(config.getString("input.trainData"), dataSetType, true);
    RegDataSet testSet = null;
    if (config.getBoolean("train.showTestProgress")) {
        testSet = TRECFormat.loadRegDataSet(config.getString("input.testData"), dataSetType, true);
    }
    LSBoost lsBoost = new LSBoost();
    RegTreeConfig regTreeConfig = new RegTreeConfig().setMaxNumLeaves(config.getInt("train.numLeaves"));
    RegTreeFactory regTreeFactory = new RegTreeFactory(regTreeConfig);
    LSBoostOptimizer optimizer = new LSBoostOptimizer(lsBoost, trainSet, regTreeFactory);
    optimizer.setShrinkage(config.getDouble("train.shrinkage"));
    optimizer.initialize();
    int progressInterval = config.getInt("train.showProgress.interval");
    int numIterations = config.getInt("train.numIterations");
    for (int i = 1; i <= numIterations; i++) {
        System.out.println("iteration " + i);
        optimizer.iterate();
        if (config.getBoolean("train.showTrainProgress") && (i % progressInterval == 0 || i == numIterations)) {
            System.out.println("training RMSE = " + RMSE.rmse(lsBoost, trainSet));
        }
        if (config.getBoolean("train.showTestProgress") && (i % progressInterval == 0 || i == numIterations)) {
            System.out.println("test RMSE = " + RMSE.rmse(lsBoost, testSet));
        }
    }
    System.out.println("training done!");
    String output = config.getString("output.folder");
    new File(output).mkdirs();
    File serializedModel = new File(output, "model");
    Serialization.serialize(lsBoost, serializedModel);
    System.out.println("model saved to " + serializedModel.getAbsolutePath());
    File reportFile = new File(output, "train_predictions.txt");
    report(lsBoost, trainSet, reportFile);
    System.out.println("predictions on the training set are written to " + reportFile.getAbsolutePath());
}