IoTDatasetGenerationFramework/src/main/java/de/tu_darmstadt/tk/SmartHomeNetworkSim/evaluation/SWCKMeansClusteringWithLabels.java

package de.tu_darmstadt.tk.SmartHomeNetworkSim.evaluation;

import java.util.Set;

import de.tu_darmstadt.tk.SmartHomeNetworkSim.core.Packet;
import weka.clusterers.SimpleKMeans;
import weka.core.Instance;
import weka.core.Instances;
import weka.core.SelectedTag;

/**
 * Unsupervised Example: K Means Clustering
 *
 * @author Andreas T. Meyer-Berg
 */
public class SWCKMeansClusteringWithLabels extends BasicPacketClassifierWitLabels {

	/**
	 * Clusterer
	 */
	private SimpleKMeans clusterer;

	/**
	 * Number of Clusters
	 */ 
	//17 works fine
	//34 found value anomalies
	protected int NUMBER_OF_CLUSTERS = Settings.NUM_CLUSTERS;
	protected double[] stdv = new double[NUMBER_OF_CLUSTERS];
	/**
	 * Initializes the k means clusterer
	 */
	public SWCKMeansClusteringWithLabels() {
		super();
		clusterer = new SimpleKMeans();
		clusterer.setSeed(42);
		//clusterer.setDisplayStdDevs(true);
		clusterer.setInitializationMethod(new SelectedTag(Settings.KMEANS_INIT,SimpleKMeans.TAGS_SELECTION));
		//clusterer.setCanopyPeriodicPruningRate(100);
		//clusterer.setCanopyT1(0.001);
		//clusterer.setCanopyT2(0.1);
		try {
			clusterer.setNumClusters(this.NUMBER_OF_CLUSTERS);
		} catch (Exception e) {
			System.out.println("Error while building cluster");
			e.printStackTrace();
		}
	}

	@Override
	public void trainModel(Instances instances) {
		try {
			clusterer.buildClusterer(instances);
			double[] sumOfSquares = new double[NUMBER_OF_CLUSTERS];
			for(Instance i: instances) {
				/**
				 * Id of the closest cluster centroid
				 */
				int x = clusterer.clusterInstance(i);
				/**
				 * centroid instance
				 */
				Instance center = clusterer.getClusterCentroids().get(x);
				/**
				 * Distance
				 */
				double dist = clusterer.getDistanceFunction().distance(center, i);
				sumOfSquares[x] += dist*dist;
			}
			/**
			 * Calculate Standard Deviations
			 */
			for(int i = 0; i<NUMBER_OF_CLUSTERS; i++)
				this.stdv[i] = Math.sqrt(sumOfSquares[i]);
		} catch (Exception e) {
			System.out.println("Failed while training the classifier");
			e.printStackTrace();
		}
	}
	private boolean test = Settings.TEST;
	
	@Override
	public double classifyInstance(Instance instance, Packet origin) throws Exception {
		/**
		 * Id of the closest cluster centroid
		 */
		int x = clusterer.clusterInstance(instance);
		/**
		 * centroid instance
		 */
		Instance center = clusterer.getClusterCentroids().get(x);
		
		double dist = clusterer.getDistanceFunction().distance(center, instance);
		if(test) { //&& dist<stdv[x] && origin.getLabel()!=0) {
			//test = false;
			System.out.println("Analysis of: "+origin.getTextualRepresentation() + " Label: " + origin.getLabel());
			System.out.println("Classified as: "+x+" Dist: "+dist+" Stdv: "+stdv[x]);
			for(int i=0; i<clusterer.getNumClusters(); i++) {
				Instance centroid = clusterer.getClusterCentroids().get(i);
				if(centroid == null)continue;
				double d = clusterer.getDistanceFunction().distance(centroid, instance);
				
				System.out.println("Cluster: "+i+" Dist: "+d+" Stdv: "+stdv[i]);
			}
			//test = false;
			System.out.println("");
		}
		if(dist < stdv[x]*Settings.ANOMALY_THRESHOLD + Settings.PRECISION_ERROR) {
			return 0;
		}
		else {
			return Double.MAX_VALUE;
		}
		
	}

	@Override
	public long getClassificationStart() {
		return 3600000;
	}

	@Override
	public String getAlgoName() {
		return "KNN with InfoLabels";
	}
}