Examples with HalfFloatMatrix org.knime.base.util.HalfFloatMatrix used on opensource projects

Example 1 with HalfFloatMatrix

use of org.knime.base.util.HalfFloatMatrix in project knime-core by knime.

the class HierarchicalClusterNodeModel method execute.

/**
 * {@inheritDoc}
 */
@Override
protected BufferedDataTable[] execute(final BufferedDataTable[] data, final ExecutionContext exec) throws Exception {
    // determine the indices of the selected columns
    List<String> inlcludedCols = m_selectedColumns.getIncludeList();
    int[] selectedColIndices = new int[inlcludedCols.size()];
    for (int count = 0; count < selectedColIndices.length; count++) {
        selectedColIndices[count] = data[0].getDataTableSpec().findColumnIndex(inlcludedCols.get(count));
    }
    BufferedDataTable inputData = data[0];
    if (inputData.size() > 65500) {
        throw new RuntimeException("At most 65,500 patterns can be clustered");
    }
    DataTable outputData = null;
    if (DistanceFunction.Names.Manhattan.toString().equals(m_distFunctionName.getStringValue())) {
        m_distFunction = ManhattanDist.MANHATTEN_DISTANCE;
    } else {
        m_distFunction = EuclideanDist.EUCLIDEAN_DISTANCE;
    }
    // generate initial clustering
    // which means that every data point is one cluster
    List<ClusterNode> clusters = initClusters(inputData, exec);
    // store the distance per each fusion step
    DataContainer fusionCont = exec.createDataContainer(createFusionSpec());
    int iterationStep = 0;
    final HalfFloatMatrix cache;
    if (m_cacheDistances.getBooleanValue()) {
        cache = new HalfFloatMatrix((int) inputData.size(), false);
        cache.fill(Float.NaN);
    } else {
        cache = null;
    }
    double max = inputData.size();
    // the number of clusters at the beginning is equal to the number
    // of data rows (each row is a cluster)
    int numberDataRows = clusters.size();
    while (clusters.size() > 1) {
        // checks if number clusters to generate output table is reached
        if (m_numClustersForOutput.getIntValue() == clusters.size()) {
            outputData = createResultTable(inputData, clusters, exec);
        }
        exec.setProgress((numberDataRows - clusters.size()) / (double) numberDataRows, clusters.size() + " clusters left to merge.");
        iterationStep++;
        exec.setProgress(iterationStep / max, "Iteration " + iterationStep + ", " + clusters.size() + " clusters remaining");
        // calculate distance between all clusters
        float currentSmallestDist = Float.MAX_VALUE;
        ClusterNode currentClosestCluster1 = null;
        ClusterNode currentClosestCluster2 = null;
        // subprogress for loop
        double availableProgress = (1.0 / numberDataRows);
        ExecutionContext subexec = exec.createSubExecutionContext(availableProgress);
        for (int i = 0; i < clusters.size(); i++) {
            exec.checkCanceled();
            ClusterNode node1 = clusters.get(i);
            for (int j = i + 1; j < clusters.size(); j++) {
                final float dist;
                ClusterNode node2 = clusters.get(j);
                // and average linkage supported.
                if (m_linkageType.getStringValue().equals(Linkage.SINGLE.name())) {
                    dist = calculateSingleLinkageDist(node1, node2, cache, selectedColIndices);
                } else if (m_linkageType.getStringValue().equals(Linkage.AVERAGE.name())) {
                    dist = calculateAverageLinkageDist(node1, node2, cache, selectedColIndices);
                } else {
                    dist = calculateCompleteLinkageDist(node1, node2, cache, selectedColIndices);
                }
                if (dist < currentSmallestDist) {
                    currentClosestCluster1 = node1;
                    currentClosestCluster2 = node2;
                    currentSmallestDist = dist;
                }
            }
        }
        subexec.setProgress(1.0);
        // make one cluster of the two closest
        ClusterNode newNode = new ClusterNode(currentClosestCluster1, currentClosestCluster2, currentSmallestDist);
        clusters.remove(currentClosestCluster1);
        clusters.remove(currentClosestCluster2);
        clusters.add(newNode);
        // store the distance per each fusion step
        fusionCont.addRowToTable(new DefaultRow(// row key
        Integer.toString(clusters.size()), // x-axis scatter plotter
        new IntCell(clusters.size()), // y-axis scatter plotter
        new DoubleCell(newNode.getDist())));
    // // print number clusters and their data points
    // LOGGER.debug("Iteration " + iterationStep + ":");
    // LOGGER.debug(" Number Clusters: " + clusters.size());
    // printClustersDataRows(clusters);
    }
    if (clusters.size() > 0) {
        m_rootNode = clusters.get(0);
    }
    fusionCont.close();
    // if there was no input data create an empty output data
    if (outputData == null) {
        outputData = createResultTable(inputData, clusters, exec);
    }
    m_dataArray = new DefaultDataArray(inputData, 1, (int) inputData.size());
    m_fusionTable = new DefaultDataArray(fusionCont.getTable(), 1, iterationStep);
    return new BufferedDataTable[] { exec.createBufferedDataTable(outputData, exec) };
}