separate the evaluation function

Author: zeyiwen

mascot/svmMain.cu

@@ -26,9 +26,10 @@ int main(int argc, char **argv)
 	argv[2] = "0.382";
 	argv[3] = "-c";
 	argv[4] = "100";
-	argv[argc - 1] = "dataset/iris.scale";
-//	argv[argc - 1] = "dataset/a1a";
-/**/
+//	argv[argc - 1] = "dataset/iris.scale";
+	argv[argc - 1] = "dataset/a1a";
+    parser.cross_validation = 0;//0 for cross-validation; 1 for training; 2 for evaluation
+
 	char fileName[1024];
 	char savedFileName[1024];
 	Parser parser;
@@ -42,18 +43,26 @@ int main(int argc, char **argv)
 
 	printf("CUDA initialized.\n");
 
-	/*if(parser.cross_validation == 1)*/
-	/*{*/
-		/*//perform cross validation*/
-		/*cout << "performing cross-validation" << endl;*/
-		/*crossValidation(parser.param, fileName);*/
-	/*}*/
-	/*else*/
+	if(parser.cross_validation == 1)
 	{
+		//perform cross validation*/
+		cout << "performing cross-validation" << endl;
+		crossValidation(parser.param, fileName);
+	}
+    else if(parser.cross_validation == 0)
+    {
 		//perform svm training
 		cout << "performing training" << endl;
-		svmModel model = trainSVM(parser.param, fileName, parser.nNumofFeature);
-	}
+		svmModel model = trainSVM(parser.param, fileName, parser.nNumofFeature);    
+    }
+    else if(parser.cross_validation == 2)
+    {
+ 		//perform svm evaluation 
+		cout << "performing evaluation" << endl;
+		svmModel model = trainSVM(parser.param, fileName, parser.nNumofFeature);    
+        evaluateSVMClassifier(model, fileName, parser.nNumofFeature);
+       
+    }
 
-	return 0;
+        return 0;
 }

mascot/trainingFunction.cu

@@ -34,7 +34,6 @@ using std::endl;
 
 void trainingByGPU(vector<vector<float_point> > &v_v_DocVector, data_info &SDataInfo, SVMParam &param);
 
-
 svmModel trainSVM(SVMParam &param, string strTrainingFileName, int nNumofFeature) {
 
     vector<vector<float_point> > v_v_DocVector;
@@ -47,20 +46,8 @@ svmModel trainSVM(SVMParam &param, string strTrainingFileName, int nNumofFeature
     rawDataRead.ReadFromFile(strTrainingFileName, nNumofFeature, v_v_DocVector, v_nLabel);
     svmProblem problem(v_v_DocVector, v_nLabel);
     svmModel model;
-    param.probability = 1;//train with probability
+    param.probability = 0;//train with probability
     model.fit(problem, param);
-    vector<int> predictLabels = model.predict(v_v_DocVector, true);
-    int numOfCorrect = 0;
-    for (int i = 0; i < v_v_DocVector.size(); ++i) {
-        if (predictLabels[i] == v_nLabel[i])
-            numOfCorrect++;
-//        for (int j = 0; j < problem.getNumOfClasses(); ++j) {
-//            printf("%.2f,",prob[i][j]);
-//        }
-//        printf("\n");
-    }
-    printf("training accuracy = %.2f%%(%d/%d)\n", numOfCorrect / (float) v_v_DocVector.size()*100, numOfCorrect,
-           (int) v_v_DocVector.size());
     return model;
 }
 
@@ -185,3 +172,30 @@ svm_model trainBinarySVM(svmProblem &problem, const SVMParam &param) {
     model.nDimension = problem.getNumOfFeatures();
     return model;
 }
+
+void evaluateSVMClassifier(svmModel &model, string strTrainingFileName, int nNumofFeature)
+{ 
+    vector<vector<float_point> > v_v_DocVector;
+    vector<int> v_nLabel;
+
+    CDataIOOps rawDataRead;
+    int nNumofInstance = 0;     //not used
+    long long nNumofValue = 0;  //not used
+    BaseLibSVMReader::GetDataInfo(strTrainingFileName, nNumofFeature, nNumofInstance, nNumofValue);
+    rawDataRead.ReadFromFile(strTrainingFileName, nNumofFeature, v_v_DocVector, v_nLabel);
+
+    //perform svm classification
+    vector<int> predictLabels = model.predict(v_v_DocVector, true);
+    int numOfCorrect = 0;
+    for (int i = 0; i < v_v_DocVector.size(); ++i) 
+    {
+        if (predictLabels[i] == v_nLabel[i])
+            numOfCorrect++;
+//        for (int j = 0; j < problem.getNumOfClasses(); ++j) {
+//            printf("%.2f,",prob[i][j]);
+//        }
+//        printf("\n");
+    }
+    printf("training accuracy = %.2f%%(%d/%d)\n", numOfCorrect / (float) v_v_DocVector.size()*100, 
+            numOfCorrect, (int) v_v_DocVector.size());
+}

mascot/trainingFunction.h

@@ -16,4 +16,6 @@ using std::string;
 svmModel trainSVM(SVMParam &param, string strTrainingFileName, int nNumofFeature);
 
 svm_model trainBinarySVM(svmProblem &problem, const SVMParam &param);
+void evaluateSVMClassifier(svmModel &model, string strTrainingFileName, int nNumofFeature);
+
 #endif /* TESTTRAINER_H_ */