/*
* RapidMiner
*
* Copyright (C) 2001-2011 by Rapid-I and the contributors
*
* Complete list of developers available at our web site:
*
* http://rapid-i.com
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see http://www.gnu.org/licenses/.
*/
package com.rapidminer.operator.clustering.clusterer;
import java.util.ArrayList;
import java.util.List;
import com.rapidminer.example.Attribute;
import com.rapidminer.example.Attributes;
import com.rapidminer.example.Example;
import com.rapidminer.example.ExampleSet;
import com.rapidminer.example.Tools;
import com.rapidminer.example.table.AttributeFactory;
import com.rapidminer.operator.OperatorCapability;
import com.rapidminer.operator.OperatorDescription;
import com.rapidminer.operator.OperatorException;
import com.rapidminer.operator.UserError;
import com.rapidminer.operator.clustering.CentroidClusterModel;
import com.rapidminer.operator.clustering.ClusterModel;
import com.rapidminer.operator.learner.CapabilityProvider;
import com.rapidminer.operator.ports.metadata.CapabilityPrecondition;
import com.rapidminer.parameter.ParameterType;
import com.rapidminer.parameter.ParameterTypeInt;
import com.rapidminer.tools.Ontology;
import com.rapidminer.tools.RandomGenerator;
import com.rapidminer.tools.math.similarity.DistanceMeasure;
import com.rapidminer.tools.math.similarity.numerical.EuclideanDistance;
/**
* This operator represents an implementation of k-means. This operator will create a cluster attribute if not present
* yet.
*
* The implementation is according to paper of C. Elkan:
* - Using the Triangle Inequality to Accelerate k-Means -
* Proceedings of the Twentieth International Conference on Machine Learning (ICML-2003), Washington DC, 2003
*
* @author Alexander Arimond
*/
public class FastKMeans extends RMAbstractClusterer implements CapabilityProvider{
/** The parameter name for "the maximal number of clusters" */
public static final String PARAMETER_K = "k";
/**
* The parameter name for "the maximal number of runs of the k method with random initialization that are
* performed"
*/
public static final String PARAMETER_MAX_RUNS = "max_runs";
/** The parameter name for "the maximal number of iterations performed for one run of the k method" */
public static final String PARAMETER_MAX_OPTIMIZATION_STEPS = "max_optimization_steps";
public FastKMeans(OperatorDescription description) {
super(description);
getExampleSetInputPort().addPrecondition(new CapabilityPrecondition(this, getExampleSetInputPort()));
}
@Override
public ClusterModel generateClusterModel(ExampleSet exampleSet) throws OperatorException {
int k = getParameterAsInt(PARAMETER_K);
int maxOptimizationSteps = getParameterAsInt(PARAMETER_MAX_OPTIMIZATION_STEPS);
int maxRuns = getParameterAsInt(PARAMETER_MAX_RUNS);
DistanceMeasure measure = new EuclideanDistance();
measure.init(exampleSet);
// checking and creating ids if necessary
Tools.checkAndCreateIds(exampleSet);
// additional checks
Tools.onlyNonMissingValues(exampleSet, "KMeans");
if (exampleSet.size() < k) {
throw new UserError(this, 142, k);
}
// extracting attribute names
Attributes attributes = exampleSet.getAttributes();
ArrayList<String> attributeNames = new ArrayList<String>(attributes.size());
for (Attribute attribute : attributes)
attributeNames.add(attribute.getName());
RandomGenerator generator = RandomGenerator.getRandomGenerator(this);
double minimalIntraClusterDistance = Double.POSITIVE_INFINITY;
CentroidClusterModel bestModel = null;
int[] bestAssignments = null;
for (int iter = 0; iter < maxRuns; iter++) {
checkForStop();
CentroidClusterModel model = new CentroidClusterModel(exampleSet, k, attributeNames, measure, getParameterAsBoolean(RMAbstractClusterer.PARAMETER_ADD_AS_LABEL), getParameterAsBoolean(RMAbstractClusterer.PARAMETER_REMOVE_UNLABELED));
// init centroids by assigning one single, unique example!
int i = 0;
for (Integer index : generator.nextIntSetWithRange(0, exampleSet.size(), k)) {
model.assignExample(i, getAsDoubleArray(exampleSet.getExample(index), attributes));
i++;
}
model.finishAssign();
// auxiliary data structures according to paper
final double [][] l = new double [exampleSet.size()][k];
final double [] u = new double [exampleSet.size()];
final boolean [] r = new boolean [exampleSet.size()];
final double [][] m_old = new double [k][attributes.size()]; // needed for step 4
final double [] s = new double [k];
final int[] centroidAssignments = new int[exampleSet.size()];
final DistanceMatrix centroidDistances = new DistanceMatrix(k);
computeClusterDistances(centroidDistances, s, model, measure);
// initialization step (has many distance calculations)
int x = 0;
for (Example example : exampleSet) {
double[] exampleValues = getAsDoubleArray(example, attributes);
double nearestDistance = measure.calculateDistance(model.getCentroidCoordinates(0), exampleValues);
l[x][0] = nearestDistance;
int nearestIndex = 0;
for (int centroidIndex = 1; centroidIndex < k; centroidIndex++) {
if (centroidDistances.get(nearestIndex, centroidIndex) >= 2 * nearestDistance) continue;
final double distance = measure.calculateDistance(model.getCentroidCoordinates(centroidIndex), exampleValues);
l[x][centroidIndex] = distance;
if (distance < nearestDistance) {
nearestDistance = distance;
nearestIndex = centroidIndex;
}
}
centroidAssignments[x] = nearestIndex;
u[x] = nearestDistance;
r[x] = false;
x++;
}
// optimization steps (repeat until convergence)
boolean stable = false;
for (int step = 0; (step < maxOptimizationSteps) && !stable; step++) {
// step 1.
computeClusterDistances(centroidDistances, s, model, measure);
int avoidedSamples = 0;
x = 0;
for (Example example : exampleSet) {
final double [] exampleValue = getAsDoubleArray(example, attributes);
// step 2.
if ( u[x] <= s[centroidAssignments[x]]){
// nothing
avoidedSamples ++;
} else {
// step 3.
for (int c = 0; c < k; c ++){
if (c != centroidAssignments[x] // (i)
&& u[x] > l[x][c] // (ii)
&& u[x] > 0.5 * centroidDistances.get(centroidAssignments[x], c) // (iii)
){
//step 3a.
final double d_x_c; // d(x,c(x))
if (r[x]){
d_x_c = measure.calculateDistance(exampleValue, model.getCentroidCoordinates(centroidAssignments[x]));
l[x][centroidAssignments[x]] = d_x_c;
u[x] = d_x_c;
r[x] = false;
} else {
d_x_c = u[x];
}
// step 3b.
if (d_x_c > l[x][c] && d_x_c > 0.5 * centroidDistances.get(centroidAssignments[x], c)){
final double d_x_c_new = measure.calculateDistance(exampleValue, model.getCentroidCoordinates(c)); // d(x,c)
l[x][c] = d_x_c_new;
if (d_x_c_new < d_x_c) {
centroidAssignments[x] = c;
u[x] = d_x_c_new;
}
}
}
}
}
model.assignExample(centroidAssignments[x], exampleValue);
x++;
}
// step 4
// first store old c
for (int c=0; c <k ; c++){
m_old[c] = model.getCentroidCoordinates(c);
}
// then compute the m(c) - here this is same as step 7
stable = model.finishAssign();
// compute all d(c,m(c))
final double [] mean_distances = new double [k];
for (int c = 0; c < k; c++){
mean_distances [c] = measure.calculateDistance(m_old[c], model.getCentroidCoordinates(c));
}
// step 5 & 6
for (x=0; x < exampleSet.size(); x++){
// step 5
for (int c = 0; c < k; c++){
final double d = l[x][c] - mean_distances[c];
if (d > 0)
l[x][c] = d;
else
l[x][c] = 0;
}
// step 6
u[x] = u[x] + mean_distances[centroidAssignments[x]];
r[x] = true;
}
}
// assessing quality of this model
double distanceSum = 0;
i = 0;
for (Example example : exampleSet) {
double distance = measure.calculateDistance(model.getCentroidCoordinates(centroidAssignments[i]), getAsDoubleArray(example, attributes));
distanceSum += distance * distance;
i++;
}
if (distanceSum < minimalIntraClusterDistance) {
bestModel = model;
minimalIntraClusterDistance = distanceSum;
bestAssignments = centroidAssignments;
}
}
bestModel.setClusterAssignments(bestAssignments, exampleSet);
if (addsClusterAttribute()) {
Attribute cluster = AttributeFactory.createAttribute("cluster", Ontology.NOMINAL);
exampleSet.getExampleTable().addAttribute(cluster);
exampleSet.getAttributes().setCluster(cluster);
int i = 0;
for (Example example : exampleSet) {
example.setValue(cluster, "cluster_" + bestAssignments[i]);
i++;
}
}
return bestModel;
}
// this is for step 1 of the paper algorithm
private void computeClusterDistances(DistanceMatrix centroidDistances, double[] s,
CentroidClusterModel model, DistanceMeasure measure) {
for (int i = 0; i < model.getNumberOfClusters(); i++){
s[i] = Double.POSITIVE_INFINITY;
}
for (int i = 0; i < model.getNumberOfClusters(); i++){
for (int j = i+1; j < model.getNumberOfClusters(); j++){
final double d = measure.calculateDistance(model.getCentroidCoordinates(i), model.getCentroidCoordinates(j));
if (d < s[i]){
s[i] = d;
}
if (d < s[j]){
s[j] = d;
}
centroidDistances.set(i, j, d);
}
}
for (int i = 0; i < model.getNumberOfClusters(); i++){
s[i] = 0.5 * s[i];
}
}
private double[] getAsDoubleArray(Example example, Attributes attributes) {
double[] values = new double[attributes.size()];
int i = 0;
for (Attribute attribute : attributes) {
values[i] = example.getValue(attribute);
i++;
}
return values;
}
@Override
public Class<? extends ClusterModel> getClusterModelClass() {
return CentroidClusterModel.class;
}
@Override
public boolean supportsCapability(OperatorCapability capability) {
switch (capability) {
case BINOMINAL_ATTRIBUTES:
case POLYNOMINAL_ATTRIBUTES:
return false;
default:
return true;
}
}
@Override
public List<ParameterType> getParameterTypes() {
List<ParameterType> types = super.getParameterTypes();
types.add(new ParameterTypeInt(PARAMETER_K, "The number of clusters which should be detected.", 2, Integer.MAX_VALUE, 2, false));
types.add(new ParameterTypeInt(PARAMETER_MAX_RUNS, "The maximal number of runs of k-Means with random initialization that are performed.", 1, Integer.MAX_VALUE, 10, false));
types.add(new ParameterTypeInt(PARAMETER_MAX_OPTIMIZATION_STEPS, "The maximal number of iterations performed for one run of k-Means.", 1, Integer.MAX_VALUE, 100, false));
types.addAll(RandomGenerator.getRandomGeneratorParameters(this));
return types;
}
}