implemented thresholds for regression models

bioclipse · Jun 20, 2012 · 1718cba · 1718cba
1 parent 31baf5d
commit 1718cba
Showing 1 changed file with 63 additions and 27 deletions.
diff --git a/plugins/net.bioclipse.ds.libsvm/src/net/bioclipse/ds/libsvm/SignaturesLibSVMPrediction.java b/plugins/net.bioclipse.ds.libsvm/src/net/bioclipse/ds/libsvm/SignaturesLibSVMPrediction.java
@@ -56,17 +56,26 @@ public class SignaturesLibSVMPrediction extends AbstractDSTest{
 	private static final String LOW_PERCENTILE = "lowPercentile";
 	private static final String TRAIN_PARAMETER = "trainFile";
 
+	private static final String REGRESSION_LOWER_THRESHOLD = "lower_threshold";
+	private static final String REGRESSION_UPPER_THRESHOLD = "upper_threshold";
+	private static final String LOW_IS_NEGATIVE = "lowIsNegative";
+
 	private static final int NR_NEAR_NEIGHBOURS = 3;
 
 
 	protected double lowPercentile;
 	protected double highPercentile;
 	protected String trainFilename;
+
+	protected Float regrUpperThreshold=null;
+	protected Float regrLowerThreshold=null;
+	protected boolean lowIsNegative=true;
+
 
     //We need to ensure that '.' is always decimal separator in all locales
     DecimalFormat formatter=new DecimalFormat("0.000");
 
-	private Vector<svm_node[]> nearNeighborData;
+//	private Vector<svm_node[]> nearNeighborData;
 
 
     //The model file
@@ -114,7 +123,7 @@ public List<String> getRequiredParameters() {
         ret.add( SIGNATURES_MIN_HEIGHT );
 //        ret.add( HIGH_PERCENTILE );
 //        ret.add( LOW_PERCENTILE );
-        ret.add( TRAIN_PARAMETER );
+//        ret.add( TRAIN_PARAMETER );
         return ret;
     }
 
@@ -152,6 +161,15 @@ public void initialize(IProgressMonitor monitor) throws DSException {
 
             System.out.println("Low percentile is: " + lowPercentile);
             System.out.println("High percentile is: " + highPercentile);
+
+        	try{
+        		regrLowerThreshold=Float.parseFloat(getParameters().get(REGRESSION_LOWER_THRESHOLD));
+        		regrUpperThreshold=Float.parseFloat(getParameters().get(REGRESSION_UPPER_THRESHOLD));
+        		lowIsNegative=Boolean.parseBoolean(getParameters().get(LOW_IS_NEGATIVE));
+        	}catch (Exception e){
+        		logger.debug(getName() + " is a regression model without thresholds");
+        	}
+
         }
 
         positiveValue=getParameters().get( "positiveValue" );
@@ -186,22 +204,22 @@ public void initialize(IProgressMonitor monitor) throws DSException {
             throw new DSException("Could not read model file '" + modelPath 
                                   + "' due to: " + e.getMessage());
         }
-
-    	String p_trainFilename = getParameters().get( TRAIN_PARAMETER );
-
-    	try {
-			trainFilename = FileUtil.getFilePath(p_trainFilename, getPluginID());
-		} catch (Exception e) {
-			e.printStackTrace();
-			throw new DSException("Error reading train file: " + e.getMessage());
-		}
-
-		// Read the train file. It will be used to retrieve near neighbors to a query from the training data.
-		try {
-			nearNeighborData = createNearNeighborData(trainFilename);
-		} catch (IOException e) {
-			returnError("could not read train file", "could not read train file");
-		}
+//
+//    	String p_trainFilename = getParameters().get( TRAIN_PARAMETER );
+//
+//    	try {
+//			trainFilename = FileUtil.getFilePath(p_trainFilename, getPluginID());
+//		} catch (Exception e) {
+//			e.printStackTrace();
+//			throw new DSException("Error reading train file: " + e.getMessage());
+//		}
+//
+//		// Read the train file. It will be used to retrieve near neighbors to a query from the training data.
+//		try {
+//			nearNeighborData = createNearNeighborData(trainFilename);
+//		} catch (IOException e) {
+//			returnError("could not read train file", "could not read train file");
+//		}
 
 
 		if (svmModel.param.svm_type == 0) // This is a classification model.
@@ -438,18 +456,23 @@ protected List<? extends ITestResult> doRunTest(ICDKMolecule cdkmol,
         int negIX = classLabels.indexOf(negativeValue);
         int predIX = classLabels.indexOf(predictedClassLabel);
 
-        if (predIX<=posIX)
+        if (predIX<=posIX){
             match.setClassification( ITestResult.POSITIVE );
-        else if (predIX>=negIX)
+        	match.setName("Result: Positive");
+        }
+        else if (predIX>=negIX){
             match.setClassification( ITestResult.NEGATIVE );
-        else
+        	match.setName("Result: Negative");
+        }
+        else{
             match.setClassification( ITestResult.INCONCLUSIVE );
+        	match.setName("Result: Inconclusive");
+        }
 
 
         if (significantSignature.length()>0){
 			//OK, color atoms
 
-        	match.setName(significantSignature);
 
         	for (int centerAtom : centerAtoms){
 
@@ -526,15 +549,28 @@ else if (predIX>=negIX)
 			}
 			System.out.println(atomGreadientComponents.toString());					
 
-
 	        ScaledResultMatch match = new ScaledResultMatch("Result: " 
                     + formatter.format( prediction ), 
-                    ITestResult.POSITIVE);
+                    ITestResult.INFORMATIVE);
 
-	        //Neg prediction means green - Negative overall results for the model
-	        if (prediction<=0)
-	        	match.setClassification(ITestResult.NEGATIVE);
+	        int result=ITestResult.INCONCLUSIVE;
+	        if (regrLowerThreshold!=null && regrUpperThreshold!=null){
+
+	        	if (prediction<regrLowerThreshold){
+	        		if (lowIsNegative)
+	        			result=ITestResult.NEGATIVE;
+	        		else
+	        			result=ITestResult.POSITIVE;
+	        	}else if (prediction>regrUpperThreshold){
+	        		if (lowIsNegative)
+	        			result=ITestResult.POSITIVE;
+	        		else
+	        			result=ITestResult.NEGATIVE;
+	        	}
+	        	match.setClassification(result);
+	        }
 
+
 	        //Color atoms according to accumulated gradient values
 	        for (int currentAtomNr : atomGreadientComponents.keySet()){
 	        	Double currentDeriv = atomGreadientComponents.get(currentAtomNr);