Convert Weka Instances to Mallet InstanceList

If you ever come across this issue the following code snippet may help you:

 /**  
    * Converts Weka Instances to Mallet InstanceList  
    * @param instances Weka instances  
    * @return Mallet instanceList  
    */  
   public static InstanceList wekaInstances2MalletInstanceList(Instances instances) {  
     Alphabet dataAlphabet = new Alphabet();  
     LabelAlphabet targetAlphabet = new LabelAlphabet();  
     InstanceList instanceList = new InstanceList(new Noop(dataAlphabet, targetAlphabet));  
     int classIndex = instances.classIndex();  
     int numAttributes = instances.numAttributes();      
     for (int i = 0; i < numAttributes; i++) {  
       if (i == classIndex) {  
         continue;  
       }  
       Attribute attribute = instances.attribute(i);  
       dataAlphabet.lookupIndex(attribute.name());        
     }  
     Attribute classAttribute = instances.attribute(classIndex);  
     int numClasses = classAttribute.numValues();      
     for (int i = 0; i < numClasses; i++) {        
       targetAlphabet.lookupLabel(classAttribute.value(i));  
     }  
     int numInstance = instances.numInstances();  
     for (int i = 0; i < numInstance; i++) {  
       weka.core.Instance instance = instances.instance(i);  
       double[] values = instance.toDoubleArray();  
       int indices[] = new int[numAttributes];  
       int count = 0;  
       for (int j = 0; j < values.length; j++) {  
         if (j != classIndex && values[j] != 0.0) {  
           values[count] = values[j];  
           indices[count] = j;  
           count++;  
         }  
       }  
       indices = Arrays.copyOf(indices, count);  
       values = Arrays.copyOf(values, count);  
       FeatureVector fv = new FeatureVector(dataAlphabet, indices, values);  
       String classValue = instance.stringValue(classIndex);  
       Label classLabel = targetAlphabet.lookupLabel(classValue);  
       Instance malletInstance = new Instance(fv, classLabel, null, null);  
       instanceList.addThruPipe(malletInstance);  
     }  
     return instanceList;  
   }  

Mallet and LibSVM

Mallet and LibSVM are the two machine learning libraries that I have been using the most. I felt the need of a way to directly use LibSVM from Mallet. As I mentioned in another post, I made a lightly refactored version of the Java implementation of LibSVM mainly for easy integration of custom kernel functions. Doing that gave me a better understanding of how LibSVM works and consequently helped me to integrate it with Mallet.

For classification tasks a Mallet instance pipe creates a FeatureVector out of an instance. So, it is quite straight forward to transform it into a format suitable for LibSVM. However, custom kernel functions that work on data structures other than vectors need to be handled differently. In the current version I have not kept any option for providing any arbitrary data structure from the Mallet end, however the code can be easily tweaked for that.

Mallet and LibSVM being separate libraries handle class labels differently. All I had to do in SVMClassifier is to align the class labels and scores from these two libraries. I have kept an option to tell LibSVM whether to predict probabilities or not which is required if you not only need the best class but also the scores given to the other classes.

If you are interested get it from github. Let me know if you have any suggestion.

Mallet and Weka

I have been using Mallet for some time now. I have also used Weka but preferred Mallet for some reasons. However, since I sometimes tend to use Weka I created a way for me to convert a Mallet InstanceList into Weka ARFF format. This also allowed me to use the classifiers in Weka quite easily.

 package ca.uwo.csd.ai.nlp.weka;  
 import cc.mallet.types.Alphabet;  
 import cc.mallet.types.FeatureVector;  
 import cc.mallet.types.Instance;  
 import cc.mallet.types.InstanceList;  
 import java.io.IOException;  
 import java.io.StringReader;  
 import weka.core.Instances;  
 /**  
  * Converts Mallet instanceList to Weka ARFF/Instances  
  * @author Syeed Ibn Faiz  
  */  
 public class Converter {  
   /**  
    * Converts Mallet InstanceList into Weka ARFF format  
    * @param instances Mallet instances  
    * @param description a String description required by Weka  
    * @return ARFF representation of the InstanceList  
    */  
   public static String convert2ARFF(InstanceList instances, String description) {  
     Alphabet dataAlphabet = instances.getDataAlphabet();  
     Alphabet targetAlphabet = instances.getTargetAlphabet();  
     StringBuilder sb = new StringBuilder();  
     sb.append("@Relation \"").append(description).append("\"\n\n");  
     int size = dataAlphabet.size();  
     for (int i = 0; i < size; i++) {  
       sb.append("@attribute \"").append(dataAlphabet.lookupObject(i).toString().replaceAll("\\s+", "_")).append("_").append(i);  
       sb.append("\" numeric\n");  
     }  
     sb.append("@attribute target {");  
     for (int i = 0; i < targetAlphabet.size(); i++) {  
       if (i != 0) sb.append(",");  
       sb.append(targetAlphabet.lookupObject(i).toString().replace(",", ";"));  
     }  
     sb.append("}\n\n@data\n");  
     for (int i = 0; i < instances.size(); i++) {  
       Instance instance = instances.get(i);  
       sb.append("{");  
       FeatureVector fv = (FeatureVector) instance.getData();  
       int[] indices = fv.getIndices();  
       double[] values = fv.getValues();  
       boolean[] attrFlag = new boolean[size];  
       double[] attrValue = new double[size];  
       for (int j = 0; j < indices.length; j++) {  
         attrFlag[indices[j]] = true;  
         attrValue[indices[j]] = values[j];  
       }        
       for (int j = 0; j < attrFlag.length; j++) {          
         if (attrFlag[j]) {            
           //sb.append(j).append(" 1, ");            
           sb.append(j).append(" ").append(attrValue[j]).append(", ");  
         }          
       }  
       sb.append(attrFlag.length).append(" ").append(instance.getTarget().toString().replace(",", ";"));  
       sb.append("}\n");        
     }  
     return sb.toString();  
   }  
   /**  
    * Converts Mallet InstanceList into Weka Instances  
    * @param instanceList  
    * @return  
    * @throws IOException   
    */  
   public static Instances convert2WekaInstances(InstanceList instanceList) throws IOException {  
     String arff = convert2ARFF(instanceList, "DESC");  
     StringReader reader = new StringReader(arff);  
     Instances instances = new Instances(reader);  
     instances.setClassIndex(instances.numAttributes() - 1);  
     return instances;  
   }  
 }  

It is now quite straight forward to call a classifier in Weka as shown in the following example:

 public static void main(String[] args) throws IOException, Exception {  
     ArrayList<Pipe> pipes = new ArrayList<Pipe>();  
     pipes.add(new Target2Label());  
     pipes.add(new CharSequence2TokenSequence());  
     pipes.add(new TokenSequence2FeatureSequence());  
     pipes.add(new FeatureSequence2FeatureVector());  
     SerialPipes pipe = new SerialPipes(pipes);  
     //prepare training instances  
     InstanceList trainingInstanceList = new InstanceList(pipe);  
     trainingInstanceList.addThruPipe(new CsvIterator(new FileReader("webkb-train-stemmed.txt"),  
         "(.*)\t(.*)", 2, 1, -1));  
     //prepare test instances  
     InstanceList testingInstanceList = new InstanceList(pipe);  
     testingInstanceList.addThruPipe(new CsvIterator(new FileReader("webkb-test-stemmed.txt"),  
         "(.*)\t(.*)", 2, 1, -1));  
     //Using a classifier in Mallet  
     ClassifierTrainer trainer = new NaiveBayesTrainer();  
     Classifier classifier = trainer.train(trainingInstanceList);  
     System.out.println("Accuracy[Mallet]: " + classifier.getAccuracy(testingInstanceList));  
     //Getting Weka Instances  
     Instances trainingInstances = Converter.convert2WekaInstances(trainingInstanceList);  
     Instances testingInstances = Converter.convert2WekaInstances(testingInstanceList);  
     //Using a classifier in Weka  
     NaiveBayesMultinomial naiveBayesMultinomial = new NaiveBayesMultinomial();  
     naiveBayesMultinomial.buildClassifier(trainingInstances);  
     Evaluation evaluation = new Evaluation(testingInstances);  
     evaluation.evaluateModel(naiveBayesMultinomial, testingInstances);  
     System.out.println("Accuracy[Weka]: " + evaluation.correct() / testingInstanceList.size());      
   }  

Using the WebKB dataset I got the following output:

 Accuracy[Mallet]: 0.836676217765043  
 Accuracy[Weka]: 0.836676217765043