public void recalculateClusterDistances(List<ClusterDendogram> clusters, Cluster.distType type) {
for (int clustOne = 0; clustOne < clusters.size(); clustOne++) {
for (int clustTwo = clustOne + 1; clustTwo < clusters.size(); clustTwo++) {
Cluster cluster_A = clusters.get(clustOne).getCluster();
Cluster cluster_B = clusters.get(clustTwo).getCluster();
cluster_A.recalculateDistanceTo(cluster_B, type, mUpperThreshold, mLowerThreshold);
cluster_B.recalculateDistanceTo(cluster_A, type, mUpperThreshold, mLowerThreshold);
}
}
}
/*
* move clusters to new cluster list, when one of the clusters that will be merged
* is found, merge it with the other cluster, then add to new cluster list
* only do this for one cluster to be merged to avoid duplicates
*/
private List<ClusterDendogram> combineClusters(
List<ClusterDendogram> clusters, Point minNdx, double correlation, Cluster.distType type) {
// System.out.printf("minNdx for cluster combining is <%f, %f>\n", minNdx.getX(),
// minNdx.getY());
ArrayList<ClusterDendogram> newClusters = new ArrayList<ClusterDendogram>();
// ArrayList<Dendogram> newDendogram = new ArrayList<Dendogram>();
for (int clusterNdx = 0; clusterNdx < clusters.size(); clusterNdx++) {
if (clusterNdx != (int) minNdx.getX() && clusterNdx != (int) minNdx.getY()) {
newClusters.add(clusters.get(clusterNdx));
// newDendogram.add(dendogram.get(clusterNdx));
} else if (clusterNdx == (int) minNdx.getX()) {
// using minNdx for cluster one for consistency and readability
// System.out.printf("minNdx X: %d minNdx Y: %d clustersLength: %d", (int) minNdx.getX(),
// (int) minNdx.getY(), clusters.size());
Cluster clusterOne = clusters.get((int) minNdx.getX()).getCluster();
Cluster clusterTwo = clusters.get((int) minNdx.getY()).getCluster();
System.out.printf(
"combining clusters:\n===\n\n%s\n and \n%s\nwith maxSimiliarty: %.03f\n\n===",
clusterOne, clusterTwo, correlation);
Cluster combinedCluster = new Cluster(clusterOne.unionWith(clusterTwo));
// using minNdx for dendogram one for consistency and readability
Dendogram leftDend = clusters.get((int) minNdx.getX()).getDendogram();
Dendogram rightDend = clusters.get((int) minNdx.getY()).getDendogram();
// Dendogram newDendogram = new DendogramNode(clusterOne.actualDistance(clusterTwo, type),
// leftDend, rightDend);
Dendogram newDendogram =
new DendogramNode(clusterOne.corrDistance(clusterTwo, type), leftDend, rightDend);
newClusters.add(new ClusterDendogram(combinedCluster, newDendogram));
}
}
// make the new cluster set into the current cluster set for next iteration
return newClusters;
// clusterer.dendogram = newDendogram;
}
// TODO check clusters vs dailyClusters, there's a problem in this method
private List<ClusterDendogram> clusterToDate(
List<ClusterDendogram> clusters,
List<ClusterDendogram> dailyClusters,
Cluster.distType type) {
// System.out.printf("Clustering clusters (%d) with new day's clusters (%d)\n", clusters.size(),
// dailyClusters.size());
// outer for loop loops over clusters in a day
// inner for loop loops over clusters built up to the current day
/*
* clustering between days uses just correlations
*/
// System.out.printf("\n***clustering to date***\n");
for (ClusterDendogram newClusterDend : dailyClusters) {
Cluster newCluster = newClusterDend.getCluster();
// double minDist = Double.MAX_VALUE;
double maxSimilarity = 0;
int closeClusterNdx = -1;
for (int clustNdx = 0; clustNdx < clusters.size(); clustNdx++) {
Cluster currClust = clusters.get(clustNdx).getCluster();
double clustDist = newCluster.corrDistance(currClust, type);
/*
System.out.printf("\n\nnewCluster: \n\t%s\n\ncurrClust:\n\t%s\n\n", newCluster, currClust);
System.out.printf("clustDist: %.03f\n", clustDist);
*/
// if (clustDist < minDist && clustDist < mLowerThreshold) {
// System.out.println("cluster to date ward's distance: " + clustDist);
// if (clustDist < minDist && clustDist >= mUpperThreshold) {
/*
System.out.printf("is %.03f > %.03f? %s\n\n", clustDist, maxSimilarity, (clustDist > maxSimilarity));
System.out.printf("mThreshold = %.03f\n", mThresholding);
*/
// if (clustDist > maxSimilarity && clustDist > mThresholding) {
if (mClusterPreference.equals("structure")) {
if (clustDist > maxSimilarity && !newCluster.isDifferent(currClust)) {
maxSimilarity = clustDist;
// System.out.printf("maxSimilarity: %.03f\n", maxSimilarity);
closeClusterNdx = clustNdx;
}
} else if (mClusterPreference.equals("similarity")) {
if (clustDist > maxSimilarity && newCluster.isSimilar(currClust)) {
maxSimilarity = clustDist;
// System.out.printf("maxSimilarity: %.03f\n", maxSimilarity);
closeClusterNdx = clustNdx;
}
}
}
// replace the cluster closest to the new Cluster with the
// oldCluster U newCluster
if (closeClusterNdx != -1) {
Cluster closeCluster = clusters.get(closeClusterNdx).getCluster();
Dendogram newDendogram =
new DendogramNode(
maxSimilarity,
newClusterDend.getDendogram(),
clusters.get(closeClusterNdx).getDendogram());
ClusterDendogram newClustDend =
new ClusterDendogram(closeCluster.unionWith(newCluster), newDendogram);
clusters.set(closeClusterNdx, newClustDend);
} else {
clusters.add(newClusterDend);
}
}
// System.out.printf("clusters combined into size %d\n", clusters.size());
// System.out.printf("\n***Finished clustering new day's clusters***\n");
return clusters;
}
private List<ClusterDendogram> clusterGroup(
List<ClusterDendogram> clusters, Cluster.distType type) {
// System.out.printf("***clustering new group***\n");
Point closeClusters = new Point(-1, -1);
// double minDist = Double.MAX_VALUE;
double maxSimilarity = 0;
boolean hasChanged;
do {
// System.out.printf("entering clustering loop...\n");
hasChanged = false;
for (int clustOne = 0; clustOne < clusters.size(); clustOne++) {
for (int clustTwo = clustOne + 1; clustTwo < clusters.size(); clustTwo++) {
Cluster cluster_A = clusters.get(clustOne).getCluster();
Cluster cluster_B = clusters.get(clustTwo).getCluster();
// System.out.printf("\n\ncluster A:\n\t%s\n\ncluster B:\n\t%s\n\n", cluster_A,
// cluster_B);
// double clustDist = cluster_A.distance(cluster_B, type);
// this will ensure that i'm only comparing based on correlations
double clustDist = cluster_A.corrDistance(cluster_B, type);
// System.out.printf("clustDist: %.03f\n", clustDist);
/*
if (clustDist > 1) {
System.err.println("cluster group clustDist: " + clustDist + " between " + cluster_A + " and " + cluster_B);
}
*/
// System.out.println("ward's distance: " + clustDist);
// if (clustDist < minDist && clustDist > 99.7 ) {
// if (clustDist < minDist && clustDist < .03 ) {
// if (clustDist < minDist && clustDist > 99.7 ) { this
// corresponds to results used in paper
// TODO investigate the results for when you use '> minDist'
// if (/*clustDist > minDist &&*/ clustDist > 99.7 ) {
// System.out.printf("clustDist: %.03f\n", clustDist);
// if (clustDist > minDist && clustDist > mThresholding ) {
// System.out.printf("is %.03f > %.03f? %s\n\n", clustDist, maxSimilarity, (clustDist >
// maxSimilarity));
// System.out.printf("mThreshold = %.03f\n", mThresholding);
// if (clustDist > maxSimilarity && clustDist > mThresholding) {
if (mClusterPreference.equals("structure")) {
if (clustDist > maxSimilarity && !cluster_A.isDifferent(cluster_B)) {
maxSimilarity = clustDist;
// System.out.printf("maxSimilarity: %.03f\n", maxSimilarity);
closeClusters = new Point(clustOne, clustTwo);
}
} else if (mClusterPreference.equals("similarity")) {
if (clustDist > maxSimilarity && cluster_A.isSimilar(cluster_B)) {
maxSimilarity = clustDist;
// System.out.printf("maxSimilarity: %.03f\n", maxSimilarity);
closeClusters = new Point(clustOne, clustTwo);
}
}
}
}
/*
* if newCluster list is a different sized then clearly two clusters were
* combined. In this case set hasChanges to true and set the cluster list to
* the new cluster list
*/
List<ClusterDendogram> newClusters =
combineClusters(clusters, closeClusters, maxSimilarity, type);
if (newClusters.size() != clusters.size()) {
hasChanged = true;
clusters = newClusters;
System.out.println("recalculating cluster distances...");
recalculateClusterDistances(clusters, type);
System.out.println("finished recalculating cluster distances...");
}
// reset various variables
closeClusters = new Point(-1, -1);
maxSimilarity = 0;
// System.out.printf("finishing clustering iteration...\n");
// continue clustering until clusters do not change
} while (hasChanged);
// System.out.printf("***Finished clustering group***\n");
return clusters;
}