public static bool trainProblem() { if (checkExistingDataset()) { SVMProblem problem = SVMProblemHelper.Load(Constants.DATA_PATH); SVMProblem randdata = SVMProblemHelper.Load(Constants.RAND_PATH); List <string> resultsstring = new List <string>(); List <SVMClass.SVMResult> ResultsList = new List <SVMClass.SVMResult>(); double C, gammasq; double Cmin = 1, Cmax = 10000, Cstep = 10; double gmin = 0.0001, gmax = 1000, gstep = 10; bool satisfied = false; while (!satisfied) { for (C = Cmin; C <= Cmax; C = C * Cstep) { for (gammasq = gmin; gammasq <= gmax; gammasq = gammasq * gstep) { SVMParameter tempparameter = new SVMParameter(); tempparameter.Type = SVMType.C_SVC; tempparameter.Kernel = SVMKernelType.RBF; tempparameter.C = C; tempparameter.Gamma = gammasq; SVMModel tempmodel = SVM.Train(problem, tempparameter); SVMProblem testData = SVMProblemHelper.Load(Constants.RAND_PATH); double[] results = testData.Predict(tempmodel); int[,] confusionMatrix; double testAccuracy = testData.EvaluateClassificationProblem(results, tempmodel.Labels, out confusionMatrix); // Do cross validation to check this parameter set is correct for the dataset or not double[] crossValidationResults; // output labels int nFold = 10; problem.CrossValidation(tempparameter, nFold, out crossValidationResults); // Evaluate the cross validation result // If it is not good enough, select the parameter set again double crossValidationAccuracy = problem.EvaluateClassificationProblem(crossValidationResults); SVMClass.SVMResult compiled = new SVMClass.SVMResult(); compiled.C = C; compiled.gamma = gammasq; compiled.testAcc = testAccuracy; compiled.crossValidAcc = crossValidationAccuracy; ResultsList.Add(compiled); } } // Evaluate the test results double maxTestAcc = ResultsList.Max(resultdata => resultdata.testAcc); //int maxTestAccIndex = ResultsList.FindIndex(resultdata => resultdata.testAcc.Equals(maxTestAcc)); double maxValidAcc = ResultsList.Max(resultdata => resultdata.crossValidAcc); //int maxValidAccIndex = ResultsList.FindIndex(resultdata => resultdata.crossValidAcc.Equals(maxValidAcc)); if (maxTestAcc < 95 || maxValidAcc < 95) { satisfied = false; Cstep--; gstep--; } else { satisfied = true; List <SVMClass.SVMResult> topResults = ResultsList.FindAll(resultdata => resultdata.testAcc.Equals(maxTestAcc)); List <SVMClass.SVMResult> topValid = ResultsList.FindAll(resultdata => resultdata.crossValidAcc.Equals(maxValidAcc)); while (topResults.Count > topValid.Count) { topResults.RemoveAt(ResultsList.FindIndex(resultsdata => resultsdata.crossValidAcc.Equals(ResultsList.Min(resultdata => resultdata.crossValidAcc)))); } double maxC = topResults.Max(resultdata => resultdata.C); int maxCIndex = topResults.FindIndex(resultdata => resultdata.C.Equals(maxC)); double bestgamma = topResults[maxCIndex].gamma; // maxC or not??? //double bestC = topResults[topResults.Count - 2].C; //topResults[maxCIndex].C; //double bestgamma = topResults[topResults.Count - 2].gamma;//topResults[maxCIndex].gamma; Console.WriteLine("Best C: " + maxC + " Best gammasq: " + bestgamma); Constants.C = maxC; Constants.gammasq = bestgamma; foreach (SVMClass.SVMResult resultdata in topResults) { Console.WriteLine(resultdata.C.ToString() + " " + resultdata.gamma.ToString()); } } } SVMParameter parameter = new SVMParameter(); parameter.Type = SVMType.C_SVC; parameter.Kernel = SVMKernelType.RBF; parameter.C = Constants.C; parameter.Gamma = Constants.gammasq; Variables.model = SVM.Train(problem, parameter); //File.WriteAllText(Constants.MODEL_PATH, String.Empty); //SVM.SaveModel(Variables.model, Constants.MODEL_PATH); Console.WriteLine("Trained and saved model.\n"); //return Variables.model; return(true); } else { MessageBox.Show("Invalid training data!"); return(false); } }
private void testSVM() { if (!holdCommandListener) { holdCommandListener = true; } string parentpath = System.AppDomain.CurrentDomain.BaseDirectory; string DATA_PATH = parentpath + "Datasets\\dataset - Copy (2).txt"; string MODEL_PATH = parentpath + "Model\\testmodel.txt"; string NEWDATA_PATH = parentpath + "Datasets\\testdata.txt"; string RESULTS_PATH = parentpath + "Datasets\\results.txt"; List <string> resultsstring = new List <string>(); SVMProblem testSet = SVMProblemHelper.Load(NEWDATA_PATH); SVMParameter testparameter = new SVMParameter(); testparameter.Type = SVMType.C_SVC; testparameter.Kernel = SVMKernelType.RBF; testparameter.C = 0.1; //Constants.C; testparameter.Gamma = 0.001; // Constants.gammasq; List <SVMClass.SVMResult> ResultsList = new List <SVMClass.SVMResult>(); SVMProblem problem = SVMProblemHelper.Load(DATA_PATH); double C = 0.001; double gammasq = 0.001; for (C = 1; C <= 1000; C = C * 10) { for (gammasq = 0.001; gammasq <= 1000; gammasq = gammasq * 10) { SVMParameter parameter = new SVMParameter(); parameter.Type = SVMType.C_SVC; parameter.Kernel = SVMKernelType.RBF; parameter.C = C; parameter.Gamma = gammasq; SVMModel model = SVM.Train(problem, parameter); //File.WriteAllText(MODEL_PATH, String.Empty); //SVM.SaveModel(model, MODEL_PATH); //Console.WriteLine("Trained and saved model.\n"); //model = SVM.LoadModel(MODEL_PATH); SVMProblem newData = SVMProblemHelper.Load(NEWDATA_PATH); //Console.Write("Predicted Result:\n"); double[] results = newData.Predict(model); //Console.Write(results[0]); int[,] confusionMatrix; double testAccuracy = newData.EvaluateClassificationProblem(results, model.Labels, out confusionMatrix); // Do cross validation to check this parameter set is correct for the dataset or not double[] crossValidationResults; // output labels int nFold = 10; problem.CrossValidation(parameter, nFold, out crossValidationResults); // Evaluate the cross validation result // If it is not good enough, select the parameter set again double crossValidationAccuracy = problem.EvaluateClassificationProblem(crossValidationResults); //Console.WriteLine("\n\nCross validation accuracy: " + crossValidationAccuracy); string temp = ""; string resultstring = "Predict accuracy: " + testAccuracy + " C: " + C + " gamma: " + gammasq + " Cross validation accuracy: " + crossValidationAccuracy; resultsstring.Add(resultstring); if (parameter.C == testparameter.C && parameter.Gamma == testparameter.Gamma) { resultsstring.Add("This one is same as separate test."); } foreach (double res in results) { temp += res.ToString() + " "; } resultsstring.Add(temp); SVMClass.SVMResult compiled = new SVMClass.SVMResult(); compiled.C = C; compiled.gamma = gammasq; compiled.testAcc = testAccuracy; compiled.crossValidAcc = crossValidationAccuracy; ResultsList.Add(compiled); } } File.WriteAllLines(RESULTS_PATH, resultsstring); SVMModel testmodel = SVM.Train(problem, testparameter); // Predict the instances in the test set double[] testResults = testSet.Predict(testmodel); foreach (double result in testResults) { Console.WriteLine(result); } // Evaluate the test results double maxTestAcc = ResultsList.Max(resultdata => resultdata.testAcc); int maxTestAccIndex = ResultsList.FindIndex(resultdata => resultdata.testAcc.Equals(maxTestAcc)); //double maxValidAcc = ResultsList.Max(resultdata => resultdata.crossValidAcc); //int maxValidAccIndex = ResultsList.FindIndex(resultdata => resultdata.crossValidAcc.Equals(maxValidAcc)); List <SVMClass.SVMResult> topResults = ResultsList.FindAll(resultdata => resultdata.testAcc.Equals(maxTestAcc)); double maxC = topResults.Max(resultdata => resultdata.C); int maxCIndex = topResults.FindIndex(resultdata => resultdata.C.Equals(maxC)); double bestC = topResults[topResults.Count - 2].C; //topResults[maxCIndex].C; double bestgamma = topResults[topResults.Count - 2].gamma; //topResults[maxCIndex].gamma; Console.WriteLine("Best C: " + bestC + " Best gammasq: " + bestgamma); foreach (SVMClass.SVMResult resultdata in topResults) { Console.WriteLine(resultdata.C.ToString() + " " + resultdata.gamma.ToString()); } //int[,] confusionMatrix; //double testAccuracy = testSet.EvaluateClassificationProblem(testResults, testmodel.Labels, out confusionMatrix); //Console.WriteLine("\n\nTest accuracy: " + testAccuracy); }