[Jboost-cvs] jboost/src/jboost/atree InstrumentedAlternatingTree.java, 1.8, 1.9

[Sunsern C. <sch...@us...>]

Update of /cvsroot/jboost/jboost/src/jboost/atree
In directory fdv4jf1.ch3.sourceforge.com:/tmp/cvs-serv31946/src/jboost/atree

Modified Files:
	InstrumentedAlternatingTree.java 
Log Message:
* improve how prediction nodes get normalized.
* multiclass support for RobustBoost

Index: InstrumentedAlternatingTree.java
===================================================================
RCS file: /cvsroot/jboost/jboost/src/jboost/atree/InstrumentedAlternatingTree.java,v
retrieving revision 1.8
retrieving revision 1.9
diff -C2 -d -r1.8 -r1.9
*** InstrumentedAlternatingTree.java	9 Feb 2009 10:47:50 -0000	1.8
--- InstrumentedAlternatingTree.java	19 Feb 2009 14:20:25 -0000	1.9
***************
*** 14,21 ****
--- 14,23 ----
  import jboost.booster.Booster;
  import jboost.booster.BrownBoost;
+ import jboost.booster.MulticlassWrapMH;
  import jboost.booster.Prediction;
  import jboost.booster.NormalizedPrediction;
  import jboost.booster.RobustBinaryPrediction;
  import jboost.booster.RobustBoost;
+ import jboost.booster.MulticlassWrapMH.MultiPrediction;
  import jboost.controller.Configuration;
  import jboost.controller.ConfigurationException;
***************
*** 389,419 ****
        // 1.a) Add the new prediction nodes to the alternating tree list
        pInt[i]= m_predictors.size();
- 
-       // RobustBoost needs to scale all of the previous
-       // hyphothesis by exp(-dt)
-       if (predictions[i] instanceof RobustBinaryPrediction) {
-         // for every RobustBinaryPrediction added before this one
-         // we scale all of them by exp(-dt)
-         
-         double dt = ((RobustBinaryPrediction)predictions[i]).getDt();
- 
-         if (dt > 0) {
-           double exp_negative_dt = Math.exp(-dt) ;
- 
-           // for each prediction before this one
-           for (int nodeidx=0;nodeidx < m_predictors.size();nodeidx++) {
-             PredictorNode cpn = (PredictorNode)m_predictors.get(nodeidx);
-             if (cpn.prediction instanceof RobustBinaryPrediction) {
- 
-               //System.out.println(">> scaling predictions");
-               //System.out.println("Before: " + ((RobustBinaryPrediction)cpn.prediction).getClassScores()[1]);
-               ((RobustBinaryPrediction)cpn.prediction).scale(exp_negative_dt);
-               //System.out.println("After: " + ((RobustBinaryPrediction)cpn.prediction).getClassScores()[1]);
- 
-             }
-           }
-         }
-       }
- 
        addPredictorNodeToList(pNode[i]);
        // 1.b) Generate the exampleMasks for the split.
--- 391,394 ----
***************
*** 464,467 ****
--- 439,510 ----
        lastBasePredictor= new AtreePredictor(splitter, parent, predictions, m_booster);
        node= findSplitter(parent, splitter);
+       
+       //--------- RobustBoost ----------//
+       
+       // binary case
+       if (m_booster instanceof RobustBoost) {
+ 
+         for (int i=0; i < predictions.length; i++) {
+         
+           // RobustBoost needs to scale all of the previous
+           // hyphothesis by exp(-dt)
+           if (predictions[i] instanceof RobustBinaryPrediction) {
+             // for every RobustBinaryPrediction added before this one
+             // we scale all of them by exp(-dt)
+             double dt = ((RobustBinaryPrediction)predictions[i]).getDt();
+             double exp_negative_dt = Math.exp(-dt) ;
+ 
+             // for each prediction before this one
+             for (int nodeidx=0;nodeidx < m_predictors.size();nodeidx++) {
+               PredictorNode cpn = (PredictorNode)m_predictors.get(nodeidx);
+               cpn.prediction.scale(exp_negative_dt);
+             }
+               
+           }
+           else {
+             throw new RuntimeException("RobustBinaryPrediction is expected. This should never happen!");
+           }
+         }
+       }
+       // multiclass case
+       else if (m_booster instanceof MulticlassWrapMH) {
+         Booster b = ((MulticlassWrapMH)m_booster).m_booster;
+         if (b instanceof RobustBoost) {
+           for (int i=0; i < predictions.length; i++) {
+ 
+             if (predictions[i] instanceof MultiPrediction) {
+ 
+               Prediction[] preds = ((MultiPrediction)predictions[i]).preds;
+               
+               for (int j=0;j<preds.length;j++) {
+               
+                 if (preds[j] instanceof RobustBinaryPrediction) {
+ 
+                   // for every RobustBinaryPrediction added before this one
+                   // we scale all of them by exp(-dt)
+                   double dt = ((RobustBinaryPrediction)preds[j]).getDt();
+                   double exp_negative_dt = Math.exp(-dt) ;
+ 
+                   // for each prediction before this one
+                   for (int nodeidx=0;nodeidx < m_predictors.size();nodeidx++) {
+                     PredictorNode cpn = (PredictorNode)m_predictors.get(nodeidx);
+                     cpn.prediction.scale(exp_negative_dt);
+                   }
+                   
+                 }
+                 else {
+                   throw new RuntimeException("RobustBinaryPrediction is expected. This should never happen!");
+                 }
+               }
+             }
+             else {
+               throw new RuntimeException("RobustBinaryPrediction is expected. This should never happen!");
+             }
+           }
+         }
+       }
+ 
+       //--------------------------------//
+         
        if (node != null && (predictions.length > 0 && !(predictions[0] instanceof NormalizedPrediction)) ) {
          for (int i=0; i < node.predictorNodes.length; i++) {
***************
*** 473,477 ****
        }
      }
!     /*
  	System.out.println("Adding Candidate: " + candidate);
  
--- 516,521 ----
        }
      }
!     
!    /*
  	System.out.println("Adding Candidate: " + candidate);
  
***************
*** 655,668 ****
  
    public boolean boosterIsFinished() {
!     if(m_booster instanceof BrownBoost){
        BrownBoost b = (BrownBoost) m_booster;
        return b.isFinished();
      }
  
!     if(m_booster instanceof RobustBoost){
        RobustBoost b = (RobustBoost) m_booster;
        return b.isFinished();
      }
! 
      double EPS = 1e-50;
      double w = m_booster.getTotalWeight();
--- 699,719 ----
  
    public boolean boosterIsFinished() {
!     if (m_booster instanceof BrownBoost){
        BrownBoost b = (BrownBoost) m_booster;
        return b.isFinished();
      }
  
!     if (m_booster instanceof RobustBoost){
        RobustBoost b = (RobustBoost) m_booster;
        return b.isFinished();
      }
!     
!     if (m_booster instanceof MulticlassWrapMH) {
!       if (((MulticlassWrapMH)m_booster).m_booster instanceof RobustBoost) {
!         RobustBoost b = (RobustBoost)(((MulticlassWrapMH)m_booster).m_booster);
!         return b.isFinished();
!       }
!     }
!     
      double EPS = 1e-50;
      double w = m_booster.getTotalWeight();