Changeset 40

Timestamp:

09/17/15 13:46:16 (9 years ago)

Author:

atrautsch

Message:

logarithmtransform -1*value bei negativen Wwerten, MedianAsReference? an Camargo Cruz Paper angepasst

Location:

trunk/CrossPare/src/de/ugoe/cs/cpdp/dataprocessing

Files:

• LogarithmTransform.java (modified) (4 diffs)
• MedianAsReference.java (modified) (6 diffs)

trunk/CrossPare/src/de/ugoe/cs/cpdp/dataprocessing/LogarithmTransform.java

@@ -35 +35 @@
                         Instance instance = testdata.instance(i);
                         for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                if( testdata.attribute(j)!=classAttribute ) {
-                                        instance.setValue(j, Math.log(1+instance.value(j)));
+                                if( testdata.attribute(j)!=classAttribute && testdata.attribute(j).isNumeric() ) {
+                                        if( instance.value(j) < 0 ) {
+                                                instance.setValue(j, (-1*(Math.log(-1*instance.value(j)))));
+                                        }else {
+                                                instance.setValue(j, Math.log(1+instance.value(j)));
+                                        }
                                 }
                         }
@@ -46 +50 @@
                                 Instance instance = traindata.instance(i);
                                 for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                        if( testdata.attribute(j)!=classAttribute ) {
-                                                instance.setValue(j, Math.log(1+instance.value(j)));
+                                        if( traindata.attribute(j)!=classAttribute && traindata.attribute(j).isNumeric() ) {
+                                                if( instance.value(j) < 0 ) {
+                                                        instance.setValue(j, (-1*(Math.log(-1*instance.value(j)))));
+                                                }else {
+                                                        instance.setValue(j, Math.log(1+instance.value(j)));
+                                                }
                                         }
                                 }
@@ -65 +73 @@
                         Instance instance = testdata.instance(i);
                         for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                if( testdata.attribute(j)!=classAttribute ) {
-                                        instance.setValue(j, Math.log(1+instance.value(j)));
+                                if( testdata.attribute(j)!=classAttribute && testdata.attribute(j).isNumeric() ) {
+                                        if( instance.value(j) < 0 ) {
+                                                instance.setValue(j, (-1*(Math.log(-1*instance.value(j)))));
+                                        }else {
+                                                instance.setValue(j, Math.log(1+instance.value(j)));
+                                        }
                                 }
                         }
@@ -75 +87 @@
                         Instance instance = traindata.instance(i);
                         for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                if( testdata.attribute(j)!=classAttribute ) {
-                                        instance.setValue(j, Math.log(1+instance.value(j)));
+                                if( traindata.attribute(j)!=classAttribute && traindata.attribute(j).isNumeric() ) {
+                                        if( instance.value(j) < 0 ) {
+                                                instance.setValue(j, (-1*(Math.log(-1*instance.value(j)))));
+                                        }else {
+                                                instance.setValue(j, Math.log(1+instance.value(j)));
+                                        }
                                 }
                         }

trunk/CrossPare/src/de/ugoe/cs/cpdp/dataprocessing/MedianAsReference.java

@@ -10 +10 @@
  * Median as reference transformation after Carmargo Cruz and Ochimizu: Towards Logistic Regression Models for Predicting Fault-prone Code across Software Projects
  * <br><br>
- * For each attribute value x, the new value is x-median of the test data
+ * For each attribute value x, the new value is x + (median of the test data - median of the current project)
  * @author Steffen Herbold
  */
@@ -30 +30 @@
         public void apply(Instances testdata, SetUniqueList<Instances> traindataSet) {
                 final Attribute classAttribute = testdata.classAttribute();
+                final double[] median = new double[testdata.numAttributes()];
                 
-                final double[] median = new double[testdata.numAttributes()];
+                // test and train have the same number of attributes
+                Attribute traindataClassAttribute;
+                double[] currentmedian = new double[testdata.numAttributes()];
                 
                 // get medians
@@ -40 +43 @@
                 }
                 
-                // update testdata
-                for( int i=0 ; i<testdata.numInstances() ; i++ ) {
-                        Instance instance = testdata.instance(i);
-                        for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                if( testdata.attribute(j)!=classAttribute ) {
-                                        instance.setValue(j, instance.value(j)-median[j]);
+                // preprocess training data
+                for( Instances traindata : traindataSet ) {
+                        // get median of current training set 
+                        traindataClassAttribute = traindata.classAttribute();
+                        for( int j=0 ; j<traindata.numAttributes() ; j++ ) {
+                                if( traindata.attribute(j)!=traindataClassAttribute && traindata.attribute(j).isNumeric()) {
+                                        currentmedian[j] = traindata.kthSmallestValue(j, (traindata.numInstances()+1)>>1); // (>>2 -> /2)
                                 }
                         }
-                }
-                
-                // preprocess training data
-                for( Instances traindata : traindataSet ) {
                         for( int i=0 ; i<traindata.numInstances() ; i++ ) {
                                 Instance instance = traindata.instance(i);
-                                for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                        if( testdata.attribute(j)!=classAttribute ) {
-                                                instance.setValue(j, instance.value(j)-median[j]);
+                                for( int j=0 ; j<traindata.numAttributes() ; j++ ) {
+                                        if( traindata.attribute(j)!=classAttribute && traindata.attribute(j).isNumeric() ) {
+                                                instance.setValue(j, instance.value(j) + (median[j] - currentmedian[j]));
                                         }
                                 }
@@ -69 +69 @@
         public void apply(Instances testdata, Instances traindata) {
                 final Attribute classAttribute = testdata.classAttribute();
-                
+                final Attribute traindataClassAttribute = traindata.classAttribute();
                 final double[] median = new double[testdata.numAttributes()];
+
+                // test and train have the same number of attributes
+                double[] currentmedian = new double[testdata.numAttributes()];
                 
                 // get medians
@@ -78 +81 @@
                         }
                 }
-                
-                // update testdata
-                for( int i=0 ; i<testdata.numInstances() ; i++ ) {
-                        Instance instance = testdata.instance(i);
-                        for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                if( testdata.attribute(j)!=classAttribute ) {
-                                        instance.setValue(j, instance.value(j)-median[j]);
-                                }
+
+                // get median of current training set 
+                for( int j=0 ; j<traindata.numAttributes() ; j++ ) {
+                        if( traindata.attribute(j)!=traindataClassAttribute && traindata.attribute(j).isNumeric() ) {
+                                currentmedian[j] = traindata.kthSmallestValue(j, (traindata.numInstances()+1)>>1); // (>>2 -> /2)
                         }
                 }
@@ -92 +92 @@
                 for( int i=0 ; i<traindata.numInstances() ; i++ ) {
                         Instance instance = traindata.instance(i);
-                        for( int j=0 ; j<testdata.numAttributes() ; j++ ) {
-                                if( testdata.attribute(j)!=classAttribute ) {
-                                        instance.setValue(j, instance.value(j)-median[j]);
+                        for( int j=0 ; j<traindata.numAttributes() ; j++ ) {
+                                if( traindata.attribute(j)!=classAttribute  && traindata.attribute(j).isNumeric() ) {
+                                        instance.setValue(j, instance.value(j) + (median[j] - currentmedian[j]));
                                 }
                         }