KNNCollectionModel added

build.properties

@@ -1,3 +1,3 @@
 extension.version=2
-extension.revision=0
-extension.update=004
+extension.revision=1
+extension.update=000

resources/com/rapidminer/resources/i18n/OperatorsDocAnomalyDetection.xml

@@ -12,7 +12,16 @@
  			algorithm proposed by Ramaswamy et al (2000) by setting the 
  			corresponding parameter. The outlier score is calculated according 
  			to the measure type selected. The higher the outlier the more 
- 			anomalous the instance is.
+ 			anomalous the instance is. 
+
+ 			The operator is also able to read and write a model containing the k
+			nearest neighbors set. Typically, 99% of the execution time is used to
+			compute the neighbors such that is a good idea to store the model, for
+			example, when looping over a parameter. The operator checks whether the
+			model and the ExampleSet fit together. The model can be used for any of
+			the nearest-neighbor based algorithms. The parameter k used to create
+			the model needs to be the same or larger as the parameter k specified in
+			the operator. Otherwise, the model is re-computed.
 			</p>
 		</help>	
 	</operator>
@@ -43,7 +52,16 @@
 			taken as the final LOF score. 
 
 			A normal instance has an outlier value of approximately 1, while 
-			outliers have values greater than 1. 			
+			outliers have values greater than 1. 	
+
+			The operator is also able to read and write a model containing the k
+			nearest neighbors set. Typically, 99% of the execution time is used to
+			compute the neighbors such that is a good idea to store the model, for
+			example, when looping over a parameter. The operator checks whether the
+			model and the ExampleSet fit together. The model can be used for any of
+			the nearest-neighbor based algorithms. The parameter k used to create
+			the model needs to be the same or larger as the parameter k specified in
+			the operator. Otherwise, the model is re-computed.		
  			</p>
 		</help>
 	</operator>
@@ -61,6 +79,15 @@
 			comparisons. LoOP is also based on the nearest neighbors set. 
 			The definition of the k-distance used is the same as the one 
 			proposed by Breunig et al [1999; 2000] to handle duplicates.
+
+			The operator is also able to read and write a model containing the k
+			nearest neighbors set. Typically, 99% of the execution time is used to
+			compute the neighbors such that is a good idea to store the model, for
+			example, when looping over a parameter. The operator checks whether the
+			model and the ExampleSet fit together. The model can be used for any of
+			the nearest-neighbor based algorithms. The parameter k used to create
+			the model needs to be the same or larger as the parameter k specified in
+			the operator. Otherwise, the model is re-computed.
  			</p>
 		</help>
 	</operator>
@@ -98,6 +125,15 @@
 			one proposed by Breunig et al [1999; 2000] to handle duplicates. The 
 			normal instances will have an outlier score of approximately 1, 
 			while outliers have a value greater than 1.  
+
+			The operator is also able to read and write a model containing the k
+			nearest neighbors set. Typically, 99% of the execution time is used to
+			compute the neighbors such that is a good idea to store the model, for
+			example, when looping over a parameter. The operator checks whether the
+			model and the ExampleSet fit together. The model can be used for any of
+			the nearest-neighbor based algorithms. The parameter k used to create
+			the model needs to be the same or larger as the parameter k specified in
+			the operator. Otherwise, the model is re-computed.
 			</p>
 		</help>
 	</operator>
@@ -114,6 +150,15 @@
 			the one proposed by  Breunig et al [1999; 2000] to handle duplicates. 
 			The normal instances will have an outlier score of approximately 1, 
 			while outliers have a value greater than 1.
+
+			The operator is also able to read and write a model containing the k
+			nearest neighbors set. Typically, 99% of the execution time is used to
+			compute the neighbors such that is a good idea to store the model, for
+			example, when looping over a parameter. The operator checks whether the
+			model and the ExampleSet fit together. The model can be used for any of
+			the nearest-neighbor based algorithms. The parameter k used to create
+			the model needs to be the same or larger as the parameter k specified in
+			the operator. Otherwise, the model is re-computed.
 			</p>
 		</help>
 	</operator>

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/COFEvaluator.java

@@ -31,12 +31,20 @@
  * 
  */
 public class COFEvaluator extends KNNEvaluator {
-
+	private int n;
+	private int k;
+	private boolean newCollection;
 	public COFEvaluator(KNNCollection knnCollection, 
 			DistanceMeasure measure, boolean parallel, int numberOfThreads, Operator logger) {
 		super(knnCollection, false, measure, parallel, numberOfThreads, logger);
 	}
-
+	public COFEvaluator(KNNCollection knnCollection, 
+			DistanceMeasure measure, boolean parallel, int numberOfThreads, Operator logger,int n, int k,boolean newCollection) {
+		super(knnCollection, false, measure, parallel, numberOfThreads, logger, n, k , newCollection);
+		this.n = n;
+		this.k = k;
+		this.newCollection = newCollection;
+	}
 	/**
 	 * 
 	 * The methods implements the COF algorithm.

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/INFLOEvaluator.java

@@ -32,10 +32,16 @@
 public class INFLOEvaluator extends KNNEvaluator {
 
 
+	private boolean newCollection;
 	public INFLOEvaluator(KNNCollection knnCollection,
 			DistanceMeasure measure, boolean parallel, int numberOfThreads, Operator logger) {
 		super(knnCollection, false, measure, parallel, numberOfThreads, logger);
 	}
+	public INFLOEvaluator(KNNCollection knnCollection,
+			DistanceMeasure measure, boolean parallel, int numberOfThreads, Operator logger,int n, int k , boolean newCollection) {
+		super(knnCollection, false, measure, parallel, numberOfThreads, logger,n,k,newCollection);
+		this.newCollection = newCollection;
+	}
 
 	public double [] evaluate() {
 

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/KNNCollection.java

@@ -19,6 +19,7 @@
  */
 package de.dfki.madm.anomalydetection.evaluator.nearest_neighbor_based;
 
+import java.io.Serializable;
 import java.util.LinkedList;
 
 /**
@@ -30,7 +31,12 @@
  * @author Mennatallah Amer
  * 
  */
-public class KNNCollection {
+public class KNNCollection implements Serializable {
+	/**
+	 *  Change this if the object changes
+	 */
+	private static final long serialVersionUID = 123456L;
+
 	/** The size of the data **/
 	int n;
 
@@ -215,5 +221,12 @@ public void updateNearestNeighbors(int point1, int point2,
 			}
 		}
 	}
-
+	public static KNNCollection clone(KNNCollection a){
+		KNNCollection ret = new KNNCollection(a.n,a.k,a.points,a.weight);
+		ret.neighborIndicies = a.neighborIndicies.clone();
+		ret.neighborDistances = a.neighborDistances.clone(); 
+		ret.numberOfNeighbors = a.numberOfNeighbors.clone();
+		ret.kdistNeighbors = a.kdistNeighbors.clone();
+			return ret;
+	}
 }

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/KNNCollectionModel.java

@@ -0,0 +1,59 @@
+/*
+ *  RapidMiner Anomaly Detection Extension
+ *
+ *  Copyright (C) 2009-2013 by Deutsches Forschungszentrum fuer
+ *  Kuenstliche Intelligenz GmbH or its licensors, as applicable.
+ *
+ *  This 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.
+ *
+ *  You should have received a copy of the GNU Affero General Public License
+ *  along with this software. If not, see <http://www.gnu.org/licenses/.
+ *
+ * Author: Johann Gebhardt
+ * Responsible: Markus Goldstein (Markus.Goldstein@dfki.de)
+ *
+ * URL: http://madm.dfki.de/rapidminer/anomalydetection
+ */ 
+package de.dfki.madm.anomalydetection.evaluator.nearest_neighbor_based;
+
+import com.rapidminer.operator.AbstractModel;
+import com.rapidminer.example.ExampleSet;
+import com.rapidminer.tools.math.similarity.DistanceMeasure;
+/**
+ * 
+ * This class is used to save the knnCollection as a RapidMiner model.
+ * 
+ * @author Johann Gebhardt
+ * 
+ */
+public class KNNCollectionModel extends AbstractModel{
+	/**
+	 *  Change this if the object changes
+	 */
+	private static final long serialVersionUID = -695692136502022L;
+	/** the saved knnCollection*/
+	KNNCollection knnCollection;
+	/** the distanceMeasure used to create the model*/
+	public DistanceMeasure measure;
+	/** returns the knnCollection*/
+	public KNNCollection get(){
+		return this.knnCollection;
+	}
+	public KNNCollectionModel(ExampleSet trainingExampleSet, KNNCollection col,DistanceMeasure measure){
+		super(trainingExampleSet);
+		this.knnCollection = col;
+		this.measure = measure;
+	}
+	public ExampleSet apply(ExampleSet tmp){
+		return tmp;
+
+	}
+	@Override
+	public String toString() {
+		return getName() + " model with k = " +knnCollection.getK();
+	}
+
+}

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/KNNEvaluator.java

@@ -62,11 +62,12 @@ public void run() {
 				for (int j = 0; j < n; j++) {
 					if (i == j)
 						continue;
+					if(newCollection){
 					double currentDistance = measure.calculateDistance(
 							knnCollection.getPoints()[i], knnCollection
 									.getPoints()[j]);
 					knnCollection.updateNearestNeighbors(i, j, currentDistance);
-
+					}
 				}
 			}
 			if (logger != null)
@@ -107,8 +108,8 @@ public void run() {
 			if (logger != null)
 				logger.logNote("Thread " + start + " " + end + " started!");
 			for (int i = start; i <= end; i++) {
-				for (int j = 0; j < i; j++) {
-
+				for (int j = 0; j <= i; j++) {
+					if(newCollection){
 					double currentDistance = measure.calculateDistance(
 							knnCollection.getPoints()[i], knnCollection
 									.getPoints()[j]);
@@ -120,6 +121,7 @@ public void run() {
 						knnCollection.updateNearestNeighbors(j, i,
 								currentDistance);
 					}
+					}
 				}
 
 			}
@@ -145,7 +147,7 @@ public void run() {
 	private Operator logger;
 	protected boolean parallel;
 	protected int numberOfThreads;
-
+	boolean newCollection = false;
 	public KNNEvaluator(KNNCollection knnCollection, boolean kth,
 			DistanceMeasure measure, boolean parallel, int numberOfThreads, Operator logger) {
 		this.knnCollection = knnCollection;
@@ -158,6 +160,19 @@ public KNNEvaluator(KNNCollection knnCollection, boolean kth,
 		this.logger = logger;
 		res = new double[n];
 	}
+	public KNNEvaluator(KNNCollection knnCollection, boolean kth,
+			DistanceMeasure measure, boolean parallel, int numberOfThreads, Operator logger,int n, int k,boolean newCollection) {
+		this.knnCollection = knnCollection;
+		this.measure = measure;
+		this.kth = kth;
+		this.n = knnCollection.getN();
+		this.k = knnCollection.getK();
+		this.parallel = parallel;
+		this.numberOfThreads = numberOfThreads;
+		this.logger = logger;
+		res = new double[n];
+		this.newCollection = newCollection;
+	}
 
 	/**
 	 * To start the evaluation process.
@@ -270,14 +285,14 @@ public void run() {
 
 	private void KNNSeq() {
 		for (int i = 0; i < n; i++) {
-			for (int j = i + 1; j < n; j++) {
-
-				double currentDistance = measure.calculateDistance(
-						knnCollection.getPoints()[i],
-						knnCollection.getPoints()[j]);
-				knnCollection.updateNearestNeighbors(i, j, currentDistance);
-				knnCollection.updateNearestNeighbors(j, i, currentDistance);
-
+			for (int j = i + 1; j <n; j++) {
+				if(newCollection) {
+					double currentDistance = measure.calculateDistance(
+							knnCollection.getPoints()[i],
+							knnCollection.getPoints()[j]);
+					knnCollection.updateNearestNeighbors(i, j, currentDistance);
+					knnCollection.updateNearestNeighbors(j, i, currentDistance);
+				}
 			}
 			setAnomalyScore(i, knnCollection.getNeighBorDistanceSoFar()[i],
 					knnCollection.getNeighBorIndiciesSoFar()[i], knnCollection

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/LOFEvaluator.java

@@ -21,7 +21,9 @@
 
 import java.util.LinkedList;
 
+import com.rapidminer.example.ExampleSet;
 import com.rapidminer.operator.Operator;
+import com.rapidminer.operator.ports.OutputPort;
 import com.rapidminer.tools.math.similarity.DistanceMeasure;
 
 /**
@@ -31,14 +33,20 @@
  * 
  */
 public class LOFEvaluator extends KNNEvaluator {
-
+	public KNNCollection savedCollection;
 	private int minK;
-
 	public LOFEvaluator(int minK, KNNCollection knnCollection,
 			DistanceMeasure measure, boolean parallel, int numberOfthreads, Operator logger) {
 		super(knnCollection, false, measure, parallel, numberOfthreads, logger);
 		this.minK = minK;
 
+	}public LOFEvaluator(int minK, KNNCollection knnCollection,
+			DistanceMeasure measure, boolean parallel, int numberOfthreads, Operator logger,int n , int k , boolean newCollection
+			) {
+		super(knnCollection, false, measure, parallel, numberOfthreads, logger,n,k,newCollection);
+		this.minK = minK;
+		this.newCollection = newCollection;
+
 	}
 	/**
 	 * The method is overridden to avoid the extra unnecessary work done
@@ -52,6 +60,7 @@ protected void setAnomalyScore(int i, double[] neighBorDistanceSoFar,
 	@Override
 	public double[] evaluate() {
 		super.evaluate();
+
 		double[] lof = lof(); 
 		return lof;
 	}
@@ -65,6 +74,7 @@ public double[] reEvaluate(int step) {
 		return lof;
 	}
 
+
 	private double [] lof(){
 		double [] lof = new double[getN()];
 		double [] lrd= new double[getN()];

src/de/dfki/madm/anomalydetection/evaluator/nearest_neighbor_based/LoOPEvaluator.java

@@ -34,13 +34,22 @@
 public class LoOPEvaluator extends KNNEvaluator {
 
 	private double lambda;
-
+	private int n;
+	private int k;
+	private boolean newCollection;
 	public LoOPEvaluator(KNNCollection knnCollection,
 			DistanceMeasure measure, double lambda, boolean parallel, int numberOfThreads, Operator logger) {
 		super(knnCollection, false, measure, parallel, numberOfThreads, logger);
 		this.lambda = lambda;
 	}
-
+	public LoOPEvaluator(KNNCollection knnCollection,
+			DistanceMeasure measure, double lambda, boolean parallel, int numberOfThreads, Operator logger,int n , int k , boolean newCollection) {
+		super(knnCollection, false, measure, parallel, numberOfThreads, logger,n ,k, newCollection);
+		this.lambda = lambda;
+		this.n = n;
+		this.k = k;
+		this.newCollection = newCollection;
+	}
 	/** The method is overridden to avoid doing extra computation **/
 	@Override
 	protected void setAnomalyScore(int i, double[] neighBorDistanceSoFar,