Merge branch 'master' of github.com:Joyeewen/gpu-svm

mascot/commandLineParser.cpp

@@ -45,10 +45,6 @@ void Parser::ParseLine(int argc, char **argv, char *pcFileName, char *pcSavedFil
 	param.weight_label = NULL;
 	param.weight = NULL;
 
-	if(argc != 6)
-	{
-		HelpInfo();
-	}
 	// parse options
 	for(i=1;i<argc;i++)
 	{
@@ -63,7 +59,20 @@ void Parser::ParseLine(int argc, char **argv, char *pcFileName, char *pcSavedFil
 			case 'c':
 				param.C = atof(argv[i]);
 				break;
-				/*
+			case 'b':
+				param.probability = atoi(argv[i]);
+				break;
+			case 'f':
+				nNumofFeature = atoi(argv[i]);
+				if(nNumofFeature < 1)
+				{
+					HelpInfo();
+				}
+				break;
+            case 'o':
+                cross_validation = atoi(argv[i]);
+                break;
+		/*
 			case 's':
 				param.svm_type = atoi(argv[i]);
 				break;
@@ -91,9 +100,6 @@ void Parser::ParseLine(int argc, char **argv, char *pcFileName, char *pcSavedFil
 			case 'h':
 				param.shrinking = atoi(argv[i]);
 				break;
-			case 'b':
-				param.probability = atoi(argv[i]);
-				break;
 			case 'q':
 				print_func = &print_null;
 				i--;
@@ -115,13 +121,6 @@ void Parser::ParseLine(int argc, char **argv, char *pcFileName, char *pcSavedFil
 				param.weight[param.nr_weight-1] = atof(argv[i]);
 				break;
 				*/
-			case 'f':
-				nNumofFeature = atoi(argv[i]);
-				if(nNumofFeature < 1)
-				{
-					HelpInfo();
-				}
-				break;
 			default:
 				fprintf(stderr,"Unknown option: -%c\n", argv[i-1][1]);
 				HelpInfo();

mascot/svmMain.cu

@@ -20,15 +20,6 @@ using std::endl;
 
 int main(int argc, char **argv)
 {
-	argc = 6;
-	argv = new char*[argc];
-	argv[1] = "-g";
-	argv[2] = "0.382";
-	argv[3] = "-c";
-	argv[4] = "100";
-	argv[argc - 1] = "dataset/iris.scale";
-//	argv[argc - 1] = "dataset/a1a";
-/**/
 	char fileName[1024];
 	char savedFileName[1024];
 	Parser parser;
@@ -42,18 +33,26 @@ 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 +46,7 @@ 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
     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 +171,30 @@ 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_ */

run.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+
+
+###options
+#svm with probability output
+PROB="-b 0" #0 for no probability output; 1 for probability output.
+
+#task type
+TASK="-o 0" #0 for training; 1 for cross validation; 2 for evaluation
+
+#gamma for RBF kernel
+GAMMA="-g 0.382"
+
+#penalty
+C="-c 100"
+
+#number of features
+NUMFEATURE="-f 123"
+
+#file name (must appear as the last argument)
+FILENAME="dataset/a1a" #"dataset/iris.scale"
+
+#print out the command before execution
+set -x
+
+#command
+./bin/mascot $PROB $TASK $GAMMA $C $NUMFEATURE $FILENAME