private boolean matrixDifferencesSmallerThanEpsilon() {
boolean areAllSmallerThanEpsilon = true;
for (int p = 0; p < numberOfPoints; p++) {
for (int c = 0; c < numberOfClusters; c++) {
double prev = prevMembershipMatrix.get(p, c);
double curr = membershipMatrix.get(p, c);
if (Math.abs(curr - prev) > epsilon) {
areAllSmallerThanEpsilon = false;
break;
}
}
}
return areAllSmallerThanEpsilon;
}
private Centroid calculateCentroidOfIndex(int index) {
double[] values = new double[pointsDimensions];
double nominator, denominator;
for (int dim = 0; dim < pointsDimensions; dim++) {
nominator = 0;
denominator = 0;
for (int i = 0; i < numberOfPoints; i++) {
nominator +=
Math.pow(membershipMatrix.get(i, index), fuzziness) * points.get(i).getDim(dim);
denominator += Math.pow(membershipMatrix.get(i, index), fuzziness);
}
values[dim] = nominator / denominator;
}
return new Centroid(values);
}
public FuzzyCMeansResults calculate() throws FileNotFoundException, UnsupportedEncodingException {
membershipMatrix = new MembershipMatrix(numberOfPoints, numberOfClusters);
List<Centroid> newCentroids = calculateNewCentroids();
List<Centroid> oldCentroids;
double exponent = 2.0 / (fuzziness - 1.0);
clock.clockStart();
for (int loop = 0; ; loop++) {
prevMembershipMatrix = membershipMatrix;
oldCentroids = newCentroids;
setInfo("[Fuzzy_C_Means] Loop " + loop);
for (int p = 0; p < numberOfPoints; p++) {
setInfo("[Fuzzy_C_Means] Loop " + loop + " | point " + p);
for (int c = 0; c < numberOfClusters; c++) {
setInfo("[Fuzzy_C_Means] Loop " + loop + " | point " + p + " | cluster " + c);
double TEMP_denominator = 0;
double point_cluster_dist = euclideanDistance(points.get(p), newCentroids.get(c));
double TEMP_nominator = Math.pow((1.0 / point_cluster_dist), exponent);
for (int r = 0; r < numberOfClusters; r++) {
double summary_point_cluster_dist =
euclideanDistance(points.get(p), newCentroids.get(r));
double ratio = Math.pow(point_cluster_dist / summary_point_cluster_dist, exponent);
TEMP_denominator += Math.pow((1.0 / summary_point_cluster_dist), exponent);
}
membershipMatrix.set(p, c, TEMP_nominator / TEMP_denominator);
}
}
newCentroids = calculateNewCentroids();
if (matrixDifferencesSmallerThanEpsilon()) {
break;
}
}
double time = clock.clockEnd();
System.out.println(newCentroids);
setInfo(
"Skin segmented. Fuzz: "
+ fuzziness
+ " Eps: "
+ epsilon
+ " Pix: "
+ numberOfPoints
+ " Time: "
+ time);
return new FuzzyCMeansResults(newCentroids, membershipMatrix);
}
