PMI Scoring now always includes the predicted class/cluster field in the output, regardless of whether probabilities are being output or not.

Author: Mark Hall

src/org/pentaho/di/trans/steps/pmi/PMIScoringData.java

@@ -468,29 +468,28 @@ public Object[][] generatePredictions( RowMetaInterface inputMeta, RowMetaInterf
 
       double[] prediction = preds[i];
 
-      if ( prediction.length == 1 || !outputProbs ) {
-        if ( supervised ) {
-          if ( classAtt.isNumeric() ) {
-            resultRow[index++] = prediction[0];
-          } else {
-            int maxProb = Utils.maxIndex( prediction );
-            if ( prediction[maxProb] > 0 ) {
-              resultRow[index++] = classAtt.value( maxProb );
-            } else {
-              resultRow[index++] =
-                  BaseMessages.getString( PMIScoringMeta.PKG, "PMIScoringData.Message.UnableToPredict" );
-            }
-          }
+      if ( supervised ) {
+        if ( classAtt.isNumeric() ) {
+          resultRow[index++] = prediction[0];
         } else {
           int maxProb = Utils.maxIndex( prediction );
           if ( prediction[maxProb] > 0 ) {
-            resultRow[index++] = maxProb;
+            resultRow[index++] = classAtt.value( maxProb );
           } else {
-            resultRow[index++] =
-                BaseMessages.getString( PMIScoringMeta.PKG, "PMIScoringData.Message.UnableToPredictCluster" );
+            resultRow[index++] = BaseMessages.getString( PMIScoringMeta.PKG, "PMIScoringData.Message.UnableToPredict" );
           }
         }
       } else {
+        int maxProb = Utils.maxIndex( prediction );
+        if ( prediction[maxProb] > 0 ) {
+          resultRow[index++] = maxProb;
+        } else {
+          resultRow[index++] =
+              BaseMessages.getString( PMIScoringMeta.PKG, "PMIScoringData.Message.UnableToPredictCluster" );
+        }
+      }
+
+      if ( ( outputProbs && classAtt == null ) || ( outputProbs && !classAtt.isNumeric() ) ) {
         // output probability distribution
         for ( double j : prediction ) {
           resultRow[index++] = j;
@@ -544,29 +543,28 @@ public Object[] generatePrediction( RowMetaInterface inputMeta, RowMetaInterface
     int index = inputMeta.size();
 
     // output for numeric class or discrete class value
-    if ( prediction.length == 1 || !outputProbs ) {
-      if ( supervised ) {
-        if ( classAtt.isNumeric() ) {
-          resultRow[index++] = prediction[0];
-        } else {
-          int maxProb = Utils.maxIndex( prediction );
-          if ( prediction[maxProb] > 0 ) {
-            resultRow[index++] = classAtt.value( maxProb );
-          } else {
-            resultRow[index++] =
-                BaseMessages.getString( PMIScoringMeta.PKG, "WekaScoringData.Message.UnableToPredict" );
-          }
-        }
+    if ( supervised ) {
+      if ( classAtt.isNumeric() ) {
+        resultRow[index++] = prediction[0];
       } else {
         int maxProb = Utils.maxIndex( prediction );
         if ( prediction[maxProb] > 0 ) {
-          resultRow[index++] = maxProb;
+          resultRow[index++] = classAtt.value( maxProb );
         } else {
-          String newVal = BaseMessages.getString( PMIScoringMeta.PKG, "PMIScoringData.Message.UnableToPredictCluster" );
-          resultRow[index++] = newVal;
+          resultRow[index++] = BaseMessages.getString( PMIScoringMeta.PKG, "WekaScoringData.Message.UnableToPredict" );
         }
       }
     } else {
+      int maxProb = Utils.maxIndex( prediction );
+      if ( prediction[maxProb] > 0 ) {
+        resultRow[index++] = maxProb;
+      } else {
+        String newVal = BaseMessages.getString( PMIScoringMeta.PKG, "PMIScoringData.Message.UnableToPredictCluster" );
+        resultRow[index++] = newVal;
+      }
+    }
+
+    if ( ( outputProbs && classAtt == null ) || ( outputProbs && !classAtt.isNumeric() ) ) {
       // output probability distribution
       for ( int i = 0; i < prediction.length; i++ ) {
         Double newVal = prediction[i];

src/org/pentaho/di/trans/steps/pmi/PMIScoringMeta.java

@@ -833,33 +833,38 @@ public void saveRep( Repository rep, ObjectId id_transformation, ObjectId id_ste
         if ( supervised ) {
           classAttName = header.classAttribute().name();
 
-          if ( header.classAttribute().isNumeric() || !m_outputProbabilities ) {
-            int
-                valueType =
-                ( header.classAttribute().isNumeric() ) ? ValueMetaInterface.TYPE_NUMBER :
-                    ValueMetaInterface.TYPE_STRING;
-
-            ValueMetaInterface newVM = ValueMetaFactory.createValueMeta( classAttName + "_predicted", valueType );
-            newVM.setOrigin( origin );
-            row.addValueMeta( newVM );
-          } else {
+          int
+              valueType =
+              ( header.classAttribute().isNumeric() ) ? ValueMetaInterface.TYPE_NUMBER : ValueMetaInterface.TYPE_STRING;
+
+          ValueMetaInterface newVM = ValueMetaFactory.createValueMeta( classAttName + "_predicted", valueType );
+          newVM.setOrigin( origin );
+          row.addValueMeta( newVM );
+
+          if ( m_outputProbabilities && !header.classAttribute().isNumeric() ) {
             for ( int i = 0; i < header.classAttribute().numValues(); i++ ) {
               String classVal = header.classAttribute().value( i );
-              ValueMetaInterface
-                  newVM =
+              // ValueMetaInterface
+              newVM =
                   ValueMetaFactory.createValueMeta( classAttName + ":" + classVal + "_predicted_prob",
                       ValueMetaInterface.TYPE_NUMBER );
               newVM.setOrigin( origin );
               row.addValueMeta( newVM );
             }
           }
         } else {
+          ValueMetaInterface
+              newVM =
+              ValueMetaFactory.createValueMeta( "cluster#_predicted", ValueMetaInterface.TYPE_NUMBER );
+          newVM.setOrigin( origin );
+          row.addValueMeta( newVM );
+
           if ( m_outputProbabilities ) {
             try {
               int numClusters = ( (PMIScoringClusterer) m_model ).numberOfClusters();
               for ( int i = 0; i < numClusters; i++ ) {
-                ValueMetaInterface
-                    newVM =
+                //  ValueMetaInterface
+                newVM =
                     ValueMetaFactory
                         .createValueMeta( "cluster_" + i + "_predicted_prob", ValueMetaInterface.TYPE_NUMBER );
                 newVM.setOrigin( origin );
@@ -869,12 +874,6 @@ public void saveRep( Repository rep, ObjectId id_transformation, ObjectId id_ste
               throw new KettleStepException(
                   BaseMessages.getString( PKG, "PMIScoringMeta.Error.UnableToGetNumberOfClusters" ), ex );
             }
-          } else {
-            ValueMetaInterface
-                newVM =
-                ValueMetaFactory.createValueMeta( "cluster#_predicted", ValueMetaInterface.TYPE_NUMBER );
-            newVM.setOrigin( origin );
-            row.addValueMeta( newVM );
           }
         }
       } catch ( KettlePluginException e ) {