Examples with Clusters edu.ucsf.rbvi.clusterMaker2.internal.algorithms.attributeClusterers.Clusters used on opensource projects

Example 61 with Clusters

use of edu.ucsf.rbvi.clusterMaker2.internal.algorithms.attributeClusterers.Clusters in project clusterMaker2 by RBVI.

the class HopachablePAM method collapse.

public Clusters collapse(int i, int j, Clusters clusters) {
    // NB    In Pollard's implementation, the choice of the new medoid probably does not change downstream results...
    Clusters c = new Clusters(clusters);
    c.merge(i, j);
    // set new cost
    switch(splitCost) {
        case AVERAGE_SILHOUETTE:
            c.setCost(1 - SilhouetteCalculator.silhouettes(this.segregations(c), c).getAverage(summarizer));
            break;
        case AVERAGE_SPLIT_SILHOUETTE:
        default:
            c.setCost(MSplitSilhouetteCalculator.averageSplitSilhouette(this, c, maxL, summarizer));
            break;
    }
    return c;
}

Example 62 with Clusters

use of edu.ucsf.rbvi.clusterMaker2.internal.algorithms.attributeClusterers.Clusters in project clusterMaker2 by RBVI.

the class HopachablePAMTest method testSplit.

@Test
public void testSplit() {
    Double[] data = { .2, .2, .8, .8, .82, .82, .4, .5, .5, .4, .15, .15, .81, .81, .14, .14, .45, .45 };
    int k = 3;
    int[] ans = { 0, 1, 1, 2, 2, 0, 1, 0, 2 };
    CyMatrix mat = CyMatrixFactory.makeSmallMatrix(9, 2, data);
    HopachablePAM pam = new HopachablePAM(null, mat, DistanceMetric.CITYBLOCK);
    pam.setParameters(9, 9, SplitCost.AVERAGE_SPLIT_SILHOUETTE, new MedianSummarizer());
    Clusters c = pam.split(false);
    // check that data are split into expected number of clusters
    assertEquals(c.getNumberOfClusters(), k);
    // check cluster assignments
    for (int i = 0; i < c.size(); ++i) {
        assertEquals(c.getClusterIndex(i), ans[i]);
    }
}

Example 63 with Clusters

use of edu.ucsf.rbvi.clusterMaker2.internal.algorithms.attributeClusterers.Clusters in project clusterMaker2 by RBVI.

the class PAMTest method testCluster.

@Test
public void testCluster() {
    System.out.println("testCluster begin");
    Double[] data = { .9, .9, .8, .8, .4, .4, .5, .5, .1, .1 };
    int k = 3;
    // {0, 0, 1, 1, 2, 2};
    int[] ans = { 0, 0, 1, 1, 2 };
    // BaseMatrix mat = new BaseMatrix(4, 2, data);
    CyMatrix mat = CyMatrixFactory.makeSmallMatrix(4, 2, data);
    PAM pam = new PAM(null, mat, DistanceMetric.CITYBLOCK);
    Clusters c = pam.cluster(k);
    System.out.println("testCluster end");
    assertEquals(c.getNumberOfClusters(), k);
    for (int i = 0; i < c.size(); ++i) {
        System.out.println("c[" + i + "] = " + c.getClusterIndex(i));
    }
    for (int i = 0; i < c.size(); ++i) {
        assertEquals(c.getClusterIndex(i), ans[i]);
    }
// NB    The current implementation fails the below test case:
// data = [.9, .9; .8, .8; .4, .4,; .5, .5; .1, .1]
// ans  = [     0,      0,      1,       1,      2]
// 
// Instead, PAM.cluster(...) yields:
// res  = [     0,      0,      1,       2,      2]
// 
// This discrepancy is due to existence of singleton clusters.
// During the build phase, element 2 is selected as a medoid, which precludes
// element 4 from becoming a medoid.
// During the swap phase, the current implementation failed to register
// the (2, 4) swap as a worthwhile swap, because the contribution to a potential
// swap is only calculated based on nonmedoids other than the candidate itself.
// There is no way to justify merging element 2 to cluster 1 and creating
// a new singleton cluster headed by element 4.
// 
// In contrast, R's cluster::pam passes this test case.
// The algorithm therein likely differs from the one upon which the current
// implementation is based.
}

Example 64 with Clusters

use of edu.ucsf.rbvi.clusterMaker2.internal.algorithms.attributeClusterers.Clusters in project clusterMaker2 by RBVI.

the class RunPAM method kcluster.

@Override
public int kcluster(int nClusters, int nIterations, CyMatrix matrix, DistanceMetric metric, int[] clusterId) {
    PAM pam = new PAM(network, matrix, metric);
    Clusters c = pam.cluster(nClusters);
    // copy results into clusterId
    for (int i = 0; i < c.size(); ++i) {
        clusterId[i] = c.getClusterIndex(i);
    }
    return c.getNumberOfClusters();
}

Example 65 with Clusters

use of edu.ucsf.rbvi.clusterMaker2.internal.algorithms.attributeClusterers.Clusters in project clusterMaker2 by RBVI.

the class MSplitSilhouetteCalculator method segregateByAverageSilhouette.

public static Clusters segregateByAverageSilhouette(Segregatable seg, int K, Summarizer summarizer) {
    Clusters split = null;
    int m = seg.size();
    // bound K
    if (K > m - 1) {
        K = m - 1;
    }
    // maximize average silhouette
    double avgSil = Double.NEGATIVE_INFINITY;
    for (int k = 2; k <= K; ++k) {
        Clusters clusters = seg.cluster(k);
        Silhouettes sils = SilhouetteCalculator.silhouettes(seg.segregations(clusters), clusters);
        double t = sils.getAverage(summarizer);
        if (t > avgSil) {
            avgSil = t;
            split = clusters;
        }
    }
    if (split != null) {
        // replace classification cost by (1 - average silhouette)
        split.setCost(1 - avgSil);
    }
    return split;
}