Fix ugly bug in the way we sent the sparse vectors to the weka-wrapper and bump version for next release.

Author: johann-petrak

creole.xml

@@ -1,7 +1,7 @@
 <!-- creole.xml for Learning_Framework -->
 <CREOLE-DIRECTORY
   ID="gate.LearningFramework"
-  VERSION="3.2"
+  VERSION="3.3"
   DESCRIPTION="Learning Framework"
   HELPURL="https://github.com/GateNLP/gateplugin-LearningFramework/wiki"
 >

src/gate/plugin/learningframework/data/CorpusRepresentationMallet.java

@@ -89,4 +89,24 @@ public void savePipe(File directory) {
 
   public abstract void addScaling(ScalingMethod scaleFeatures);
 
+  // TODO: need to do this better: make sure if there are thousands of 
+  // features that we only show a subset and the number?
+  /*
+  public String toString() {
+    Alphabet dataAlph = pipe.getDataAlphabet();
+    Alphabet targetAlph = pipe.getTargetAlphabet();
+    StringBuilder sb = new StringBuilder();
+    sb.append("CorpusRepresentationMallet{dataalphabet=");
+    sb.append(dataAlph.toString());
+    sb.append(",targetalphabet=");
+    if(targetAlph==null) {
+      sb.append("null");
+    } else {
+      sb.append(targetAlph.toString());
+    }
+    sb.append("}");
+    return sb.toString();
+  }
+  */
+  
 }

src/gate/plugin/learningframework/data/CorpusRepresentationMalletTarget.java

@@ -70,7 +70,7 @@ public CorpusRepresentationMalletTarget(FeatureInfo fi, ScalingMethod sm, Target
   }
 
   /**
-   * Create a new instance based on the pipe stored in directory.
+   * Create a new CRMT instance based on the pipe stored in directory.
    * @param directory
    * @return 
    */

src/gate/plugin/learningframework/engines/Engine.java

@@ -77,6 +77,7 @@ public static Engine loadEngine(File directory, String parms) {
     // representation if necessary.
     eng.loadModel(directory, parms);
     eng.loadMalletCorpusRepresentation(directory);
+    //System.err.println("Loaded mallet corpus representation: "+eng.getCorpusRepresentationMallet());
 
     // we could stop growh right after loading, but that would interfere with engines which
     // allow updating, incremental learning etc. 

src/gate/plugin/learningframework/engines/EngineWekaExternal.java

@@ -211,7 +211,7 @@ public EvaluationResult evaluate(String algorithmParameters, EvaluationMethod ev
   public List<GateClassification> classify(AnnotationSet instanceAS, AnnotationSet inputAS, AnnotationSet sequenceAS, String parms) {
     CorpusRepresentationMalletTarget data = (CorpusRepresentationMalletTarget)corpusRepresentationMallet;
     data.stopGrowth();
-    System.err.println("Running EngineWeka.classify on document "+instanceAS.getDocument().getName());
+    //System.err.println("Running EngineWeka.classify on document "+instanceAS.getDocument().getName());
     List<GateClassification> gcs = new ArrayList<GateClassification>();
     LFPipe pipe = (LFPipe)data.getRepresentationMallet().getPipe();
     for(Annotation instAnn : instanceAS.inDocumentOrder()) {
@@ -230,8 +230,8 @@ public List<GateClassification> classify(AnnotationSet instanceAS, AnnotationSet
       int[] locations = sdv.getLocations();
       double[] values = sdv.getValues();
       for(int i=0;i<locs;i++) {
-        locations[i] = fv.location(i);
-        values[i] = fv.value(i);
+        locations[i] = fv.indexAtLocation(i);
+        values[i] = fv.value(locations[i]);
       }
       // send the vector over to the weka process
       process.writeObject(sdv);
@@ -246,7 +246,7 @@ public List<GateClassification> classify(AnnotationSet instanceAS, AnnotationSet
         // this is an error, lets panic for now
         throw new RuntimeException("Got a response from the Weka process which is not double[] but "+obj.getClass());
       }
-      System.err.println("Sent vector: locs/values="+Arrays.toString(locations)+"/"+Arrays.toString(values)+", ret="+Arrays.toString(ret));
+      //System.err.println("Sent vector: locs/values="+Arrays.toString(locations)+"/"+Arrays.toString(values)+", ret="+Arrays.toString(ret));
       GateClassification gc = null;
       // now check if the mallet representation and the weka process agree 
       // on if we have regression or classification

src/gate/plugin/learningframework/export/CorpusExporterMalletSeq.java

@@ -5,20 +5,10 @@
  */
 package gate.plugin.learningframework.export;
 
-import cc.mallet.pipe.Pipe;
-import cc.mallet.types.Alphabet;
-import cc.mallet.types.FeatureVector;
-import cc.mallet.types.Instance;
 import cc.mallet.types.InstanceList;
-import gate.plugin.learningframework.Globals;
 import gate.plugin.learningframework.data.CorpusRepresentationMallet;
 import gate.plugin.learningframework.engines.Info;
-import gate.plugin.learningframework.features.Datatype;
-import gate.plugin.learningframework.features.FeatureExtraction;
 import java.io.File;
-import java.io.FileNotFoundException;
-import java.io.FileOutputStream;
-import java.io.PrintStream;
 
 /**
  *