/*

  *   This program is free software: you can redistribute it and/or modify

  *   it under the terms of the GNU 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 General Public License for more details.

  *

  *   You should have received a copy of the GNU General Public License

  *   along with this program.  If not, see <http://www.gnu.org/licenses/>.

  */

 /*

  *    LWL.java

  *    Copyright (C) 1999-2012 University of Waikato, Hamilton, New Zealand

  *

  */

 package weka.classifiers.lazy;

 import java.util.Enumeration;

 import java.util.Vector;

 import weka.classifiers.Classifier;

 import weka.classifiers.SingleClassifierEnhancer;

 import weka.classifiers.UpdateableClassifier;

 import weka.core.Capabilities;

 import weka.core.Capabilities.Capability;

 import weka.core.Instance;

 import weka.core.Instances;

 import weka.core.Option;

 import weka.core.RevisionUtils;

 import weka.core.TechnicalInformation;

 import weka.core.TechnicalInformation.Field;

 import weka.core.TechnicalInformation.Type;

 import weka.core.TechnicalInformationHandler;

 import weka.core.Utils;

 import weka.core.WeightedInstancesHandler;

 import weka.core.neighboursearch.LinearNNSearch;

 import weka.core.neighboursearch.NearestNeighbourSearch;

 /**

  <!-- globalinfo-start -->

  * Locally weighted learning. Uses an instance-based algorithm to assign instance weights which are then used by a specified WeightedInstancesHandler.<br/>

  * Can do classification (e.g. using naive Bayes) or regression (e.g. using linear regression).<br/>

  * <br/>

  * For more info, see<br/>

  * <br/>

  * Eibe Frank, Mark Hall, Bernhard Pfahringer: Locally Weighted Naive Bayes. In: 19th Conference in Uncertainty in Artificial Intelligence, 249-256, 2003.<br/>

  * <br/>

  * C. Atkeson, A. Moore, S. Schaal (1996). Locally weighted learning. AI Review..

  * <p/>

  <!-- globalinfo-end -->

  *

  <!-- technical-bibtex-start -->

  * BibTeX:

  * <pre>

  * @inproceedings{Frank2003,

  *    author = {Eibe Frank and Mark Hall and Bernhard Pfahringer},

  *    booktitle = {19th Conference in Uncertainty in Artificial Intelligence},

  *    pages = {249-256},

  *    publisher = {Morgan Kaufmann},

  *    title = {Locally Weighted Naive Bayes},

  *    year = {2003}

  * }

  * 

  * @article{Atkeson1996,

  *    author = {C. Atkeson and A. Moore and S. Schaal},

  *    journal = {AI Review},

  *    title = {Locally weighted learning},

  *    year = {1996}

  * }

  * </pre>

  * <p/>

  <!-- technical-bibtex-end -->

  *

  <!-- options-start -->

  * Valid options are: <p/>

  * 

  * <pre> -A

  *  The nearest neighbour search algorithm to use (default: weka.core.neighboursearch.LinearNNSearch).

  * </pre>

  * 

  * <pre> -K &lt;number of neighbours&gt;

  *  Set the number of neighbours used to set the kernel bandwidth.

  *  (default all)</pre>

  * 

  * <pre> -U &lt;number of weighting method&gt;

  *  Set the weighting kernel shape to use. 0=Linear, 1=Epanechnikov,

  *  2=Tricube, 3=Inverse, 4=Gaussian.

  *  (default 0 = Linear)</pre>

  * 

  * <pre> -D

  *  If set, classifier is run in debug mode and

  *  may output additional info to the console</pre>

  * 

  * <pre> -W

  *  Full name of base classifier.

  *  (default: weka.classifiers.trees.DecisionStump)</pre>

  * 

  * <pre> 

  * Options specific to classifier weka.classifiers.trees.DecisionStump:

  * </pre>

  * 

  * <pre> -D

  *  If set, classifier is run in debug mode and

  *  may output additional info to the console</pre>

  * 

  <!-- options-end -->

  *

  * @author Len Trigg (trigg@cs.waikato.ac.nz)

  * @author Eibe Frank (eibe@cs.waikato.ac.nz)

  * @author Ashraf M. Kibriya (amk14[at-the-rate]cs[dot]waikato[dot]ac[dot]nz)

  * @version $Revision: 8034 $ 

  */

 public class LWL 

   extends SingleClassifierEnhancer

   implements UpdateableClassifier, WeightedInstancesHandler, 

              TechnicalInformationHandler {

   /** for serialization. */

   static final long serialVersionUID = 1979797405383665815L;

   /** The training instances used for classification. */

   protected Instances m_Train;

   /** The number of neighbours used to select the kernel bandwidth. */

   protected int m_kNN = -1;

   /** The weighting kernel method currently selected. */

   protected int m_WeightKernel = LINEAR;

   /** True if m_kNN should be set to all instances. */

   protected boolean m_UseAllK = true;

   /** The nearest neighbour search algorithm to use. 

    * (Default: weka.core.neighboursearch.LinearNNSearch) 

    */

   protected NearestNeighbourSearch m_NNSearch =  new LinearNNSearch();

   /** The available kernel weighting methods. */

   public static final int LINEAR       = 0;

   public static final int EPANECHNIKOV = 1;

   public static final int TRICUBE      = 2;  

   public static final int INVERSE      = 3;

   public static final int GAUSS        = 4;

   public static final int CONSTANT     = 5;

   /** a ZeroR model in case no model can be built from the data. */

   protected Classifier m_ZeroR;

   /**

    * Returns a string describing classifier.

    * @return a description suitable for

    * displaying in the explorer/experimenter gui

    */

   public String globalInfo() {

     return 

         "Locally weighted learning. Uses an instance-based algorithm to "

       + "assign instance weights which are then used by a specified "

       + "WeightedInstancesHandler.\n"

       + "Can do classification (e.g. using naive Bayes) or regression "

       + "(e.g. using linear regression).\n\n"

       + "For more info, see\n\n"

       + getTechnicalInformation().toString();

   }

   /**

    * Returns an instance of a TechnicalInformation object, containing 

    * detailed information about the technical background of this class,

    * e.g., paper reference or book this class is based on.

    * 

    * @return the technical information about this class

    */

   public TechnicalInformation getTechnicalInformation() {

     TechnicalInformation         result;

     TechnicalInformation         additional;

     result = new TechnicalInformation(Type.INPROCEEDINGS);

     result.setValue(Field.AUTHOR, "Eibe Frank and Mark Hall and Bernhard Pfahringer");

     result.setValue(Field.YEAR, "2003");

     result.setValue(Field.TITLE, "Locally Weighted Naive Bayes");

     result.setValue(Field.BOOKTITLE, "19th Conference in Uncertainty in Artificial Intelligence");

     result.setValue(Field.PAGES, "249-256");

     result.setValue(Field.PUBLISHER, "Morgan Kaufmann");

     additional = result.add(Type.ARTICLE);

     additional.setValue(Field.AUTHOR, "C. Atkeson and A. Moore and S. Schaal");

     additional.setValue(Field.YEAR, "1996");

     additional.setValue(Field.TITLE, "Locally weighted learning");

     additional.setValue(Field.JOURNAL, "AI Review");

     return result;

   }

   /**

    * Constructor.

    */

   public LWL() {    

     m_Classifier = new weka.classifiers.trees.DecisionStump();

   }

   /**

    * String describing default classifier.

    * 

    * @return the default classifier classname

    */

   protected String defaultClassifierString() {

     return "weka.classifiers.trees.DecisionStump";

   }

   /**

    * Returns an enumeration of the additional measure names 

    * produced by the neighbour search algorithm.

    * @return an enumeration of the measure names

    */

   public Enumeration enumerateMeasures() {

     return m_NNSearch.enumerateMeasures();

   }

   /**

    * Returns the value of the named measure from the 

    * neighbour search algorithm.

    * @param additionalMeasureName the name of the measure to query for its value

    * @return the value of the named measure

    * @throws IllegalArgumentException if the named measure is not supported

    */

   public double getMeasure(String additionalMeasureName) {

     return m_NNSearch.getMeasure(additionalMeasureName);

   }

   /**

    * Returns an enumeration describing the available options.

    *

    * @return an enumeration of all the available options.

    */

   public Enumeration listOptions() {

     Vector newVector = new Vector(3);

     newVector.addElement(new Option("\tThe nearest neighbour search " +

                                     "algorithm to use " +

                                     "(default: weka.core.neighboursearch.LinearNNSearch).\n",

                                     "A", 0, "-A"));

     newVector.addElement(new Option("\tSet the number of neighbours used to set"

                                     +" the kernel bandwidth.\n"

                                     +"\t(default all)",

                                     "K", 1, "-K <number of neighbours>"));

     newVector.addElement(new Option("\tSet the weighting kernel shape to use."

                                     +" 0=Linear, 1=Epanechnikov,\n"

                                     +"\t2=Tricube, 3=Inverse, 4=Gaussian.\n"

                                     +"\t(default 0 = Linear)",

                                     "U", 1,"-U <number of weighting method>"));

     Enumeration enu = super.listOptions();

     while (enu.hasMoreElements()) {

       newVector.addElement(enu.nextElement());

     }

     return newVector.elements();

   }

   /**

    * Parses a given list of options. <p/>

    *

    <!-- options-start -->

    * Valid options are: <p/>

    * 

    * <pre> -A

    *  The nearest neighbour search algorithm to use (default: weka.core.neighboursearch.LinearNNSearch).

    * </pre>

    * 

    * <pre> -K &lt;number of neighbours&gt;

    *  Set the number of neighbours used to set the kernel bandwidth.

    *  (default all)</pre>

    * 

    * <pre> -U &lt;number of weighting method&gt;

    *  Set the weighting kernel shape to use. 0=Linear, 1=Epanechnikov,

    *  2=Tricube, 3=Inverse, 4=Gaussian.

    *  (default 0 = Linear)</pre>

    * 

    * <pre> -D

    *  If set, classifier is run in debug mode and

    *  may output additional info to the console</pre>

    * 

    * <pre> -W

    *  Full name of base classifier.

    *  (default: weka.classifiers.trees.DecisionStump)</pre>

    * 

    * <pre> 

    * Options specific to classifier weka.classifiers.trees.DecisionStump:

    * </pre>

    * 

    * <pre> -D

    *  If set, classifier is run in debug mode and

    *  may output additional info to the console</pre>

    * 

    <!-- options-end -->

    *

    * @param options the list of options as an array of strings

    * @throws Exception if an option is not supported

    */

   public void setOptions(String[] options) throws Exception {

     String knnString = Utils.getOption('K', options);

     if (knnString.length() != 0) {

       setKNN(Integer.parseInt(knnString));

     } else {

       setKNN(-1);

     }

     String weightString = Utils.getOption('U', options);

     if (weightString.length() != 0) {

       setWeightingKernel(Integer.parseInt(weightString));

     } else {

       setWeightingKernel(LINEAR);

     }

     String nnSearchClass = Utils.getOption('A', options);

     if(nnSearchClass.length() != 0) {

       String nnSearchClassSpec[] = Utils.splitOptions(nnSearchClass);

       if(nnSearchClassSpec.length == 0) { 

         throw new Exception("Invalid NearestNeighbourSearch algorithm " +

                             "specification string."); 

       }

       String className = nnSearchClassSpec[0];

       nnSearchClassSpec[0] = "";

       setNearestNeighbourSearchAlgorithm( (NearestNeighbourSearch)

                   Utils.forName( NearestNeighbourSearch.class, 

                                  className, 

                                  nnSearchClassSpec)

                                         );

     }

     else 

       this.setNearestNeighbourSearchAlgorithm(new LinearNNSearch());

     super.setOptions(options);

   }

   /**

    * Gets the current settings of the classifier.

    *

    * @return an array of strings suitable for passing to setOptions

    */

   public String [] getOptions() {

     String [] superOptions = super.getOptions();

     String [] options = new String [superOptions.length + 6];

     int current = 0;

     options[current++] = "-U"; options[current++] = "" + getWeightingKernel();

     if ( (getKNN() == 0) && m_UseAllK) {

       options[current++] = "-K"; options[current++] = "-1";

     }

     else {

       options[current++] = "-K"; options[current++] = "" + getKNN();

     }

     options[current++] = "-A";

     options[current++] = m_NNSearch.getClass().getName()+" "+Utils.joinOptions(m_NNSearch.getOptions()); 

     System.arraycopy(superOptions, 0, options, current,

                      superOptions.length);

     return options;

   }

   /**

    * Returns the tip text for this property.

    * @return tip text for this property suitable for

    * displaying in the explorer/experimenter gui

    */

   public String KNNTipText() {

     return "How many neighbours are used to determine the width of the "

       + "weighting function (<= 0 means all neighbours).";

   }

   /**

    * Sets the number of neighbours used for kernel bandwidth setting.

    * The bandwidth is taken as the distance to the kth neighbour.

    *

    * @param knn the number of neighbours included inside the kernel

    * bandwidth, or 0 to specify using all neighbors.

    */

   public void setKNN(int knn) {

     m_kNN = knn;

     if (knn <= 0) {

       m_kNN = 0;

       m_UseAllK = true;

     } else {

       m_UseAllK = false;

     }

   }

   /**

    * Gets the number of neighbours used for kernel bandwidth setting.

    * The bandwidth is taken as the distance to the kth neighbour.

    *

    * @return the number of neighbours included inside the kernel

    * bandwidth, or 0 for all neighbours

    */

   public int getKNN() {

     return m_kNN;

   }

   /**

    * Returns the tip text for this property.

    * @return tip text for this property suitable for

    * displaying in the explorer/experimenter gui

    */

   public String weightingKernelTipText() {

     return "Determines weighting function. [0 = Linear, 1 = Epnechnikov,"+

            "2 = Tricube, 3 = Inverse, 4 = Gaussian and 5 = Constant. "+

            "(default 0 = Linear)].";

   }

   /**

    * Sets the kernel weighting method to use. Must be one of LINEAR, 

    * EPANECHNIKOV,  TRICUBE, INVERSE, GAUSS or CONSTANT, other values

    * are ignored.

    *

    * @param kernel the new kernel method to use. Must be one of LINEAR,

    * EPANECHNIKOV,  TRICUBE, INVERSE, GAUSS or CONSTANT.

    */

   public void setWeightingKernel(int kernel) {

     if ((kernel != LINEAR)

         && (kernel != EPANECHNIKOV)

         && (kernel != TRICUBE)

         && (kernel != INVERSE)

         && (kernel != GAUSS)

         && (kernel != CONSTANT)) {

       return;

     }

     m_WeightKernel = kernel;

   }

   /**

    * Gets the kernel weighting method to use.

    *

    * @return the new kernel method to use. Will be one of LINEAR,

    * EPANECHNIKOV,  TRICUBE, INVERSE, GAUSS or CONSTANT.

    */

   public int getWeightingKernel() {

     return m_WeightKernel;

   }

   /**

    * Returns the tip text for this property.

    * @return tip text for this property suitable for

    * displaying in the explorer/experimenter gui

    */

   public String nearestNeighbourSearchAlgorithmTipText() {

     return "The nearest neighbour search algorithm to use (Default: LinearNN).";

   }

   /**

    * Returns the current nearestNeighbourSearch algorithm in use.

    * @return the NearestNeighbourSearch algorithm currently in use.

    */

   public NearestNeighbourSearch getNearestNeighbourSearchAlgorithm() {

     return m_NNSearch;

   }

   /**

    * Sets the nearestNeighbourSearch algorithm to be used for finding nearest

    * neighbour(s).

    * @param nearestNeighbourSearchAlgorithm - The NearestNeighbourSearch class.

    */

   public void setNearestNeighbourSearchAlgorithm(NearestNeighbourSearch nearestNeighbourSearchAlgorithm) {

     m_NNSearch = nearestNeighbourSearchAlgorithm;

   }

   /**

    * Returns default capabilities of the classifier.

    *

    * @return      the capabilities of this classifier

    */

   public Capabilities getCapabilities() {

     Capabilities      result;

     if (m_Classifier != null) {

       result = m_Classifier.getCapabilities();

     } else {

       result = super.getCapabilities();

     }

     result.setMinimumNumberInstances(0);

     // set dependencies

     for (Capability cap: Capability.values())

       result.enableDependency(cap);

     return result;

   }

   /**

    * Generates the classifier.

    *

    * @param instances set of instances serving as training data 

    * @throws Exception if the classifier has not been generated successfully

    */

   public void buildClassifier(Instances instances) throws Exception {

     if (!(m_Classifier instanceof WeightedInstancesHandler)) {

       throw new IllegalArgumentException("Classifier must be a "

                                          + "WeightedInstancesHandler!");

     }

     // can classifier handle the data?

     getCapabilities().testWithFail(instances);

     // remove instances with missing class

     instances = new Instances(instances);

     instances.deleteWithMissingClass();

     // only class? -> build ZeroR model

     if (instances.numAttributes() == 1) {

       System.err.println(

           "Cannot build model (only class attribute present in data!), "

           + "using ZeroR model instead!");

       m_ZeroR = new weka.classifiers.rules.ZeroR();

       m_ZeroR.buildClassifier(instances);

       return;

     }

     else {

       m_ZeroR = null;

     }

     m_Train = new Instances(instances, 0, instances.numInstances());

     m_NNSearch.setInstances(m_Train);

   }

   /**

    * Adds the supplied instance to the training set.

    *

    * @param instance the instance to add

    * @throws Exception if instance could not be incorporated

    * successfully

    */

   public void updateClassifier(Instance instance) throws Exception {

     if (m_Train == null) {

       throw new Exception("No training instance structure set!");

     }

     else if (m_Train.equalHeaders(instance.dataset()) == false) {

       throw new Exception("Incompatible instance types\n" + m_Train.equalHeadersMsg(instance.dataset()));

     }

     if (!instance.classIsMissing()) {

       m_NNSearch.update(instance);

       m_Train.add(instance);

     }

   }

   /**

    * Calculates the class membership probabilities for the given test instance.

    *

    * @param instance the instance to be classified

    * @return preedicted class probability distribution

    * @throws Exception if distribution can't be computed successfully

    */

   public double[] distributionForInstance(Instance instance) throws Exception {

     // default model?

     if (m_ZeroR != null) {

       return m_ZeroR.distributionForInstance(instance);

     }

     if (m_Train.numInstances() == 0) {

       throw new Exception("No training instances!");

     }

     m_NNSearch.addInstanceInfo(instance);

     int k = m_Train.numInstances();

     if( (!m_UseAllK && (m_kNN < k)) /*&&

        !(m_WeightKernel==INVERSE ||

          m_WeightKernel==GAUSS)*/ ) {

       k = m_kNN;

     }

     Instances neighbours = m_NNSearch.kNearestNeighbours(instance, k);

     double distances[] = m_NNSearch.getDistances();

     if (m_Debug) {

       System.out.println("Test Instance: "+instance);

       System.out.println("For "+k+" kept " + neighbours.numInstances() + " out of " + 

                          m_Train.numInstances() + " instances.");

     }

     //IF LinearNN has skipped so much that <k neighbours are remaining.

     if(k>distances.length)

       k = distances.length;

     if (m_Debug) {

       System.out.println("Instance Distances");

       for (int i = 0; i < distances.length; i++) {

         System.out.println("" + distances[i]);

       }

     }

     // Determine the bandwidth

     double bandwidth = distances[k-1];

     // Check for bandwidth zero

     if (bandwidth <= 0) {

       //if the kth distance is zero than give all instances the same weight

       for(int i=0; i < distances.length; i++)

         distances[i] = 1;

     } else {

       // Rescale the distances by the bandwidth

       for (int i = 0; i < distances.length; i++)

         distances[i] = distances[i] / bandwidth;

     }

     // Pass the distances through a weighting kernel

     for (int i = 0; i < distances.length; i++) {

       switch (m_WeightKernel) {

         case LINEAR:

           distances[i] = 1.0001 - distances[i];

           break;

         case EPANECHNIKOV:

           distances[i] = 3/4D*(1.0001 - distances[i]*distances[i]);

           break;

         case TRICUBE:

           distances[i] = Math.pow( (1.0001 - Math.pow(distances[i], 3)), 3 );

           break;

         case CONSTANT:

           //System.err.println("using constant kernel");

           distances[i] = 1;

           break;

         case INVERSE:

           distances[i] = 1.0 / (1.0 + distances[i]);

           break;

         case GAUSS:

           distances[i] = Math.exp(-distances[i] * distances[i]);

           break;

       }

     }

     if (m_Debug) {

       System.out.println("Instance Weights");

       for (int i = 0; i < distances.length; i++) {

         System.out.println("" + distances[i]);

       }

     }

     // Set the weights on the training data

     double sumOfWeights = 0, newSumOfWeights = 0;

     for (int i = 0; i < distances.length; i++) {

       double weight = distances[i];

       Instance inst = (Instance) neighbours.instance(i);

       sumOfWeights += inst.weight();

       newSumOfWeights += inst.weight() * weight;

       inst.setWeight(inst.weight() * weight);

       //weightedTrain.add(newInst);

     }

     // Rescale weights

     for (int i = 0; i < neighbours.numInstances(); i++) {

       Instance inst = neighbours.instance(i);

       inst.setWeight(inst.weight() * sumOfWeights / newSumOfWeights);

     }

     // Create a weighted classifier

     m_Classifier.buildClassifier(neighbours);

     if (m_Debug) {

       System.out.println("Classifying test instance: " + instance);

       System.out.println("Built base classifier:\n" 

                          + m_Classifier.toString());

     }

     // Return the classifier's predictions

     return m_Classifier.distributionForInstance(instance);

   }

   /**

    * Returns a description of this classifier.

    *

    * @return a description of this classifier as a string.

    */

   public String toString() {

     // only ZeroR model?

     if (m_ZeroR != null) {

       StringBuffer buf = new StringBuffer();

       buf.append(this.getClass().getName().replaceAll(".*\\.", "") + "\n");

       buf.append(this.getClass().getName().replaceAll(".*\\.", "").replaceAll(".", "=") + "\n\n");

       buf.append("Warning: No model could be built, hence ZeroR model is used:\n\n");

       buf.append(m_ZeroR.toString());

       return buf.toString();

     }

     if (m_Train == null) {

       return "Locally weighted learning: No model built yet.";

     }

     String result = "Locally weighted learning\n"

       + "===========================\n";

     result += "Using classifier: " + m_Classifier.getClass().getName() + "\n";

     switch (m_WeightKernel) {

     case LINEAR:

       result += "Using linear weighting kernels\n";

       break;

     case EPANECHNIKOV:

       result += "Using epanechnikov weighting kernels\n";

       break;

     case TRICUBE:

       result += "Using tricube weighting kernels\n";

       break;

     case INVERSE:

       result += "Using inverse-distance weighting kernels\n";

       break;

     case GAUSS:

       result += "Using gaussian weighting kernels\n";

       break;

     case CONSTANT:

       result += "Using constant weighting kernels\n";

       break;

     }

     result += "Using " + (m_UseAllK ? "all" : "" + m_kNN) + " neighbours";

     return result;

   }

   /**

    * Returns the revision string.

    * 

    * @return                the revision

    */

   public String getRevision() {

     return RevisionUtils.extract("$Revision: 8034 $");

   }

   /**

    * Main method for testing this class.

    *

    * @param argv the options

    */

   public static void main(String [] argv) {

     runClassifier(new LWL(), argv);

   }

 }