public override ClassifierResult Build()
{
//For this example (and indeed, many scenarios), the default
//parameters will suffice.
Parameter parameters = new Parameter();
double C;
double Gamma;
//This will do a grid optimization to find the best parameters
//and store them in C and Gamma, outputting the entire
//search to params.txt.
ParameterSelection.Grid(_trainProblem, parameters, "params.txt", out C, out Gamma);
parameters.C = C;
parameters.Gamma = Gamma;
//Train the model using the optimal parameters.
_model = Training.Train(_trainProblem, parameters);
// пробегаем по всем клиентски данным и сохраняем результат
var probDictList = new Dictionary <string, Dictionary <int, double> >();
foreach (string id in _testDataDict.Keys)
{
if (!probDictList.ContainsKey(id))
{
probDictList.Add(id, new Dictionary <int, double>());
}
foreach (var sarr in _testDataDict[id])
{
var y = PredictProba(sarr);
double prob = y.Probs[1];
int kmax = probDictList[id].Keys.Count == 0 ? 0 : probDictList[id].Keys.Max() + 1;
probDictList[id].Add(kmax, prob);
}
}
// вероятность дефолта определяется как среднее по записям для клиента
var probDict = new Dictionary <string, double>();
foreach (var id in probDictList.Keys)
{
int cnt = probDictList[id].Keys.Count();
double prob = 0;
foreach (var d in probDictList[id].Keys)
{
prob += probDictList[id][d];
}
if (!probDict.ContainsKey(id))
{
probDict.Add(id, prob / cnt);
}
}
// находим статистики классификации
var rlist = new RocItem[_resultDict.Count]; // массив для оценки результата
int idx = 0;
foreach (string id in probDict.Keys)
{
if (rlist[idx] == null)
{
rlist[idx] = new RocItem();
}
rlist[idx].Prob = probDict[id]; // среднее по наблюдениям
rlist[idx].Target = _resultDict[id];
rlist[idx].Predicted = 0;
idx++;
}
Array.Sort(rlist, (o1, o2) => (1 - o1.Prob).CompareTo(1 - o2.Prob));
var cres = ResultCalc.GetResult(rlist, 0.05);
var clsRes = new ClassifierResult();
clsRes.BestResult = cres;
clsRes.LastResult = cres;
clsRes.ResDict.Add(0, cres);
return(clsRes);
}
public void svmproject(Dictionary <int, Characteristic> Characteristics)
{
Dictionary <int, Characteristic> _characteristics = new Dictionary <int, Characteristic>();
_characteristics = Characteristics;
Model model;
RangeTransform range;
double C;
double gamma;
// default values
Parameter parameters = new Parameter();
//parameters.SvmType = SvmType.C_SVC;
//parameters.KernelType = KernelType.RBF;
//parameters.Degree = 3;
//parameters.Gamma = 0;
//parameters.Coefficient0 = 0;
//parameters.Nu = 0.5;
//parameters.CacheSize = 40;
//parameters.C = 1000;
//parameters.EPS = 1e-3;
//parameters.P = 0.1;
//parameters.Shrinking = true;
//parameters.WeightCount = 0;
//parameters.WeightLabels = new int[0];
//parameters.Weights = new double[0];
//parameters.C = 5;
//parameters.Gamma = 1;
string str = null;
for (int i = 1; i < _characteristics.Count(); i++)
{
if (_characteristics[i].IsQualifiedColony == false && _characteristics[i].IsInvalidColony == true)
{
str = str + "0";
str = str + " 1:" + _characteristics[i].Area.ToString() + " 2:" + _characteristics[i].MajToMinAxisRatio.ToString() +
" 3:" + _characteristics[i].CentreAcerageColor.R.ToString() + " 4:" + _characteristics[i].CentreAcerageColor.G.ToString()
+ " 5:" + _characteristics[i].CentreAcerageColor.B.ToString() + "\r\n";
}
if (_characteristics[i].IsQualifiedColony == true && _characteristics[i].IsInvalidColony == false)
{
str = str + "1";
str = str + " 1:" + _characteristics[i].Area.ToString() + " 2:" + _characteristics[i].MajToMinAxisRatio.ToString() +
" 3:" + _characteristics[i].CentreAcerageColor.R.ToString() + " 4:" + _characteristics[i].CentreAcerageColor.G.ToString()
+ " 5:" + _characteristics[i].CentreAcerageColor.B.ToString() + "\r\n";
}
}
if (str != null)
{
byte[] array = Encoding.ASCII.GetBytes(str);
MemoryStream stream = new MemoryStream(array); //convert stream 2 string
Problem train = new Problem();
train = Problem.Read(stream);
range = Scaling.DetermineRange(train);
train = Scaling.Scale(train, range);
//String outfile001="D:\\parameters.txt";
ParameterSelection.Grid(train, parameters, @"D:\\parameters.txt", out C, out gamma);
parameters.C = C;
parameters.Gamma = gamma;
model = Training.Train(train, parameters);
//MessageBox.Show("学习完毕");
//stream.Dispose();
stream.Close();
}
else
{
MessageBox.Show("无学习数据");
model = null;
range = null;
}
string str1 = null;
for (int i = 1; i < _characteristics.Count(); i++)
{
str1 = str1 + "0";
str1 = str1 + " 1:" + _characteristics[i].Area.ToString() + " 2:" + _characteristics[i].MajToMinAxisRatio.ToString() +
" 3:" + _characteristics[i].CentreAcerageColor.R.ToString() + " 4:" + _characteristics[i].CentreAcerageColor.G.ToString()
+ " 5:" + _characteristics[i].CentreAcerageColor.B.ToString() + "\r\n";
}
if (str1 != null)
{
byte[] array = Encoding.ASCII.GetBytes(str1);
MemoryStream stream = new MemoryStream(array); //convert stream 2 string
Problem pre = new Problem();
pre = Problem.Read(stream);
pre = Scaling.Scale(pre, range);
Prediction.Predict(pre, @"D:\result.txt", model, false);
MessageBox.Show("筛选完毕");
//stream.Dispose();
stream.Close();
}
else
{
MessageBox.Show("无筛选数据");
}
//svm_problem prob = new svm_problem();
//prob.l = point_list.Count;
//prob.y = new double[prob.l];
// if(param.svm_type == svm_parameter.EPSILON_SVR ||
// param.svm_type == svm_parameter.NU_SVR)
//{
// if(param.gamma == 0) param.gamma = 1;
// prob.x = new svm_node[prob.l][];
// for(int i=0;i<prob.l;i++)
// {
// point p = (point)point_list[i];
// prob.x[i][0] = new svm_node();
// prob.x[i][0].index = 1;
// prob.x[i][0].value_Renamed = p.x;
// prob.y[i] = p.y;
// }
// svm_model model = svm.svm_train(prob, param);
// svm_node[] x = new svm_node[1];
// x[0] = new svm_node();
// x[0].index = 1;
// int[] j = new int[XLEN];
//C = Convert.ToInt16(numericUpDown8.Value);
//gamma = Convert.ToInt16(numericUpDown9.Value);
//StudyAlgorithm study = new StudyAlgorithm();
//study.GetModel(AllColony, C, gamma, out model, out range);
//ScreenAlgorithm screenAlgorithm = new ScreenAlgorithm();
//screenAlgorithm.ScreenTheColony(CharacteristicsValue, model, range);
}
static void Main(string[] args)
{
SpamDSClass sds = new SpamDSClass();
FireFlyAlgorithm ff = new FireFlyAlgorithm();
ParameterSelection ps = new ParameterSelection();
//reading the phishing emails and the test Datas
string testFileFolderName = "TrainTestDataSpam";
string trainTestDataPath = String.Format(Environment.CurrentDirectory + "\\{0}", testFileFolderName); //filepath for the training and test dataset
string[] trainTestFileURL = System.IO.Directory.GetFileSystemEntries(trainTestDataPath);
string extractedVectorsFolderName = "ExtractedFeatures";
string extractedVectorsFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", extractedVectorsFolderName);
string[] evFileURL = System.IO.Directory.GetFileSystemEntries(extractedVectorsFilePath);
string outputEvaluationFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "OutputEvaluations.txt");
string outpuVectorFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "OutputVectors.txt");
PDSClass pds = new PDSClass();
//string SVAccuracyFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "SVAccuracy.txt");
/**
* //extract and format spam words
* string spamWordsFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "StopWords.txt");
* sds.ExtractFormatSpamWords(spamWordsFilePath);
**/
/**
* //Extract UCI dataset
* string UCIInputFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "SpamDataset.txt");
* sds.ExtractFormatUCIDataset(UCIInputFilePath);
* string UCIOutputFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "SpamDatasetOutput.txt");
* sds.ExtractVectorUCIDataset(UCIOutputFilePath);
***/
/**
* //Extract UCI Card dataset
* string outputFileName = "ZooOutput.txt";
* string CreditCardInputFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "Zoo.txt");
* sds.ExtractFormatUCIDataset(CreditCardInputFilePath, outputFileName); //extract, format and write vectors to file
* string UCIOutputFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", outputFileName);
* sds.ExtractVectorUCIDataset(UCIOutputFilePath, outputFileName);
***/
/**
* //extract and format credit card dataset
* string outputFileName = "CreditCardDSOutput.txt";
* //string CreditCardInputFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "CreditCardDS.txt");
* //sds.ExtractFormatUCIDataset(CreditCardInputFilePath, outputFileName); //extract, format and write vectors to file
* string OutputFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", outputFileName);
* sds.ExtractVectorCreditCard(OutputFilePath, outputFileName);
* //sds.ReadVectorFromFile(OutputFilePath); //read vector values from file
**/
File.WriteAllText(ps.SVAccuracyFilePath, string.Empty);
File.WriteAllText(outputEvaluationFilePath, string.Empty); //deleting the contents of the file holding the results for new results
File.WriteAllText(ps.filePath, string.Empty); //deleting the contents of the file holding the extracted C,Gamma and CV values
File.WriteAllText(sds.extractedFeaturesFilePathTrainDS, string.Empty);
File.WriteAllText(ps.filePath2, string.Empty);
File.WriteAllText(ps.filePath3, string.Empty);
//File.WriteAllText(outpuVectorFilePath, string.Empty); //deleting the contents of the file holding the results for new results
//int NumofFeatures = 17;
double DatasetEntropy = 0.0;
int TotalMails = 0;
//string[] features = sds.Features(); //saving the features in the string array
int[,] HamSpamCount = new int[, ] {
}; //count for the total number of ham and phish emails
string[] testTrainFolderFiles = new string[] { };
string folderFiles;
//double[] informationGain = new double[NumofFeatures];
//Dictionary<string, double> feat_infoGain = sds.Feature_InfoGain(features, informationGain); //saving the information gain for all the individual features
Dictionary <string, double> feat_infoGain = new Dictionary <string, double>(); //saving the information gain for all the individual features
double classficatnAccuracySum = 0.0; //cummulative classification accuracy
double classficatnAccuracy = 0; //classification accuracy for each iteration
DateTime start = DateTime.Now;
double totalPhishing = 0, totalHam = 0; int mailGrandTotal = 0;
//initializing all the variables that will used for evaluating the performance of the classifier
double FP = 0.0, FN = 0.0, TP = 0.0, TN = 0.0, P = 0.0, F_M = 0.0, sumFP = 0.0, sumF_M = 0.0, sumFN = 0.0, sumTP = 0.0, sumTN = 0.0, sumP = 0.0, sumTime = 0.0;
double storagePercentage = 0.0, sumStoragePercentage = 0.0;
int N_Fold = 10; //number of folds
int n_Runs = 1; //number of runs
double[,] NormalizedVector = new double[, ] {
}; //normalized vector values for each features
Program p = new Program();
double avgRuns = 0.0, avgFP = 0.0, avgFN = 0.0, avgR = 0.0, avgPr = 0.0, avgFM = 0.0, avgTime = 0.0, avgStoragePercentage = 0.0; //avg=> average
double globalBest = double.MinValue;
//change the boolean value appropriately to choose the task you want to perform (either vector value extraction or email classification)
//Note: Both values must not be true! This is to reduce processing time
bool extractVectorValues = false; //switch for extract
bool Emailclassification = true; //switch for classification
double C = new double();
double Gamma = new double();
//double[] CV = new double[2];
List <double> CV1 = new List <double>(); //Save the CV accuracy, C and Gamma for comparison
List <double> CV2 = new List <double>(); //Save the CV accuracy, C and Gamma for comparison
List <int[, ]> vectorList = new List <int[, ]>(); //save the list of vectors for each fold
List <Dictionary <string, int> > trainMailList = new List <Dictionary <string, int> >(); //save training mails for each fold
List <Dictionary <string, int> > testMailList = new List <Dictionary <string, int> >(); //save test emails for each fold
List <string[]> featList = new List <string[]>();
int NumofFeatures = 0;
int[,] vector = new int[, ] {
};
Dictionary <string, int> trainMail_Class = new Dictionary <string, int>(); //variable containing the emails and classes of all the training emails
Dictionary <string, int> testMail_Class = new Dictionary <string, int>(); //variable containing the emails and classes of all the test emails
string[] features = new string[] { };
List <Dictionary <string, double> > feat_infoGainList = new List <Dictionary <string, double> >();
//for (int aa = 0; aa < n_Runs; aa++)
//{
// classficatnAccuracySum = 0.0;
// sumFP = 0.0;
// sumTP = 0.0; //Recall
// sumFN = 0.0;
// sumF_M = 0.0;
// sumP = 0.0;
// sumTime = 0.0;
//for (int a = 0; a < N_Fold; a++)
//{
if (extractVectorValues == true) //if the value of ExtractVectorValues is true, only extract email vector values and dont perform classification
{
for (int a = 0; a < N_Fold; a++)
{
n_Runs = 1; // change number of runs from its default value (i.e 10) to 1 (to avoid repeating the extraction process 10 times) since we wanna do is to extract the vector values from each emails
string[] trainFileURLs = new string[] { }; //urls for the train emails (i.e. the emails used for training the classifier)
string[] testFileURLs = new string[] { }; //urls for the test emails (i.e. the emails used for testing the classifier)
trainMail_Class = new Dictionary <string, int>();
testMail_Class = new Dictionary <string, int>();
string[] trainMailFileNames = new string[trainMail_Class.Count]; //the file names for all the emails in the training dataset
string[] testMailFileNames = new string[] { }; //the file names for all the emails in the test dataset
string trainMailLabel; int spamCount = 0, hamCount = 0; double phishPercentage, hamPercentage;
features = new string[] { };
//processing the training dataset for the each fold
for (int i = 0; i < trainTestFileURL.Length; i++)
{
if (i.Equals(a))
{
continue; //skipping one email folder, which is to be used for testing the trained classifier (i.e. the current test dataset)
}
testTrainFolderFiles = System.IO.Directory.GetFiles(trainTestFileURL[i]); //getting the filenames of all the emails in the training dataset
trainFileURLs = trainFileURLs.Concat(testTrainFolderFiles).ToArray(); //get all the urls for the test emails
trainMailFileNames = sds.getFileNames(trainFileURLs); //get the file names for all the test mails
for (int j = 0; j < testTrainFolderFiles.Length; j++) //processing all the emails in the current training dataset for classification
{
trainMailLabel = trainMailFileNames[j].Substring(0, 2); //getting the label for each email, HM(for Ham Mails) and PM(for Phish Mails)
//folderFiles = File.ReadAllText(testTrainFolderFiles[j]); //extracting the content of each email in each email folder
//trainMail_Class[sds.ProcessMails(folderFiles)] = (trainMailLabel.Equals("SM")) ? 1 : 0; //processing each email and assigning label to the emails based on the folders each emails come from.
trainMail_Class[testTrainFolderFiles[j]] = (trainMailLabel.Equals("SM")) ? 1 : 0; //processing each email and assigning label to the emails based on the folders each emails come from.
if (trainMail_Class.ElementAt(j).Value == 1)
{
spamCount++; //counting the total number of ham and phishing to get their percentage
}
else
{
hamCount++;
}
}
}
//processing the test dataset for each fold
for (int i = a; i < a + 1; i++)
{
testTrainFolderFiles = System.IO.Directory.GetFiles(trainTestFileURL[i]);
testFileURLs = testFileURLs.Concat(testTrainFolderFiles).ToArray();
testMailFileNames = sds.getFileNames(testFileURLs);
for (int j = 0; j < testTrainFolderFiles.Length; j++)
{
trainMailLabel = testMailFileNames[j].Substring(0, 2);
//folderFiles = File.ReadAllText(testTrainFolderFiles[j]);
//testMail_Class[sds.ProcessMails(folderFiles)] = (trainMailLabel.Equals("SM")) ? 1 : 0; //processing each email and assigning label to the emails based on the folders each emails come from.
testMail_Class[testTrainFolderFiles[j]] = (trainMailLabel.Equals("SM")) ? 1 : 0; //processing each email and assigning label to the emails based on the folders each emails come from.
if (testMail_Class.ElementAt(j).Value == 1)
{
spamCount++;
}
else
{
hamCount++;
}
}
}
//calculating the percentage of spam and ham email in the dataset
phishPercentage = (double)spamCount / (double)(trainMail_Class.Count + testMail_Class.Count);
hamPercentage = (double)hamCount / (double)(trainMail_Class.Count + testMail_Class.Count);
mailGrandTotal = spamCount + hamCount;
totalHam = hamCount; totalPhishing = spamCount;
//Information Gain
features = pds.Features();
NumofFeatures = features.Count();
TotalMails = trainMail_Class.Count;
vector = new int[TotalMails, NumofFeatures];
double[] informationGainn = new double[NumofFeatures];
pds.processVector(vector, trainMail_Class, features, trainFileURLs, NumofFeatures); //extracting the vector values of all the features
int[,] HamPhishCount = new int[NumofFeatures, 4];
pds.FeatureVectorSum(NumofFeatures, TotalMails, vector, trainMail_Class, HamPhishCount); // calculating the total number of zeros and ones for both phishing and ham emails
DatasetEntropy = pds.Entropy(trainMail_Class); //calculating the entropy for the entire dataset
pds.CalInformationGain(NumofFeatures, HamPhishCount, informationGainn, TotalMails, DatasetEntropy); //calculating information gain for each feature
feat_infoGain = pds.Feature_InfoGain(features, informationGainn); //assisgning the calculated information gain to each feature
//process vector for training Dataset
int NumofFeatures2 = NumofFeatures - 9;
string[] newFeatures = new string[NumofFeatures2];
for (int i = 0; i < NumofFeatures2; i++)
{
newFeatures[i] = feat_infoGain.ElementAt(i).Key; //copying the best 8 features with the highest information gain
}
vector = new int[TotalMails, NumofFeatures2];
pds.processVector(vector, trainMail_Class, newFeatures, trainFileURLs, NumofFeatures2);
//extract vectors of the training data
pds.extractVectors(vector, trainMail_Class, NumofFeatures2, "trainingDS", a);
//process vector for testing Dataset
TotalMails = testMail_Class.Count;
vector = new int[TotalMails, NumofFeatures2];
pds.processVector(vector, testMail_Class, newFeatures, testFileURLs, NumofFeatures2);
NormalizedVector = ff.Normalize(vector); //normalize the all vector values for test data
//extract vectors of the test data
pds.extractVectors(vector, testMail_Class, NumofFeatures2, "testDS", a);
/***
* vector = sds.processTrainVector(trainMail_Class, ref features); //extracting the vector values of all the features
*
* TotalMails = trainMail_Class.Count;
* NumofFeatures = vector.GetLength(1);
* double[] informationGain = new double[NumofFeatures];
* HamSpamCount = new int[NumofFeatures, 4];
* sds.FeatureVectorSum(NumofFeatures, TotalMails, vector, trainMail_Class, HamSpamCount); // calculating the total number of zeros and ones for both phishing and ham emails
* DatasetEntropy = sds.Entropy(trainMail_Class, NumofFeatures); //calculating the entropy for the entire dataset
*
* sds.CalInformationGain(NumofFeatures, HamSpamCount, informationGain, TotalMails, DatasetEntropy);//calculating information gain for each feature
* feat_infoGain = sds.Feature_InfoGain(features, informationGain, NumofFeatures); //assisgning the calculated information gain to each feature
*
* //extract features with information gain greater than zero
* List<string> topFeat = new List<string>();
* foreach (KeyValuePair<string, double> feat in feat_infoGain)
* {
* if (feat.Value > 0)
* topFeat.Add(feat.Key);
* }
*
* NumofFeatures = topFeat.Count; //update number of features to number of features with high information gain
* vector = sds.processVector(trainMail_Class, topFeat.ToArray(), NumofFeatures); //extracting the vector values of all the features
*
* vectorList.Add(vector); //save vectors for each fold
* trainMailList.Add(trainMail_Class); //save training mails for each fold
* testMailList.Add(testMail_Class); //save test mails for each fold
* featList.Add(topFeat.ToArray());
* feat_infoGainList.Add(feat_infoGain);
* }
*
*
* //get the feature with the lowest count
* NumofFeatures = vectorList[0].GetLength(1); //get the number of features for each fold
* foreach (int[,] vec in vectorList)
* {
* int NumofFeat = vec.GetLength(1);
* if (NumofFeat < NumofFeatures)
* NumofFeatures = NumofFeat; //get the feature with the lowest count
* }
*
* if (NumofFeatures >= 40) //retrict number of features to 100, if number of selected features is greater than 100
* NumofFeatures = 40;
*
* //For each fold, process the extracted vectors
* for (int a = 0; a < N_Fold; a++)
* {
* //extract vectors of the training data
* sds.extractVectors(vectorList[a], trainMailList[a], NumofFeatures, "trainingDS", a);
*
* //process vector for testing Dataset
* TotalMails = testMailList[a].Count;
* vector = new int[TotalMails, NumofFeatures];
* vector = sds.processVector(testMailList[a], featList[a], NumofFeatures); //extracting the vector values of all the features
* sds.extractVectors(vector, testMailList[a], NumofFeatures, "testDS", a); //extract vectors of the test data
*
* if (a.Equals(9))
* {
* Console.Write("Extraction Completed....");
* Console.ReadKey();
* }
***/
}
}
else //perform email classification - don't extract!
{
n_Runs = 1;
for (int r = 0; r < n_Runs; r++)
{
classficatnAccuracySum = 0.0; sumFP = 0.0; sumTP = 0.0; sumFN = 0.0; sumF_M = 0.0; sumP = 0.0; sumTime = 0.0; sumStoragePercentage = 0.0;
for (int a = 0; a < N_Fold; a++)
{
if (Emailclassification == true) //if the value of EmailClassification is true, perform email classification and dont extract emails
{
//SVM Classfication begins here
DateTime start1 = DateTime.Now;
//First, read in the training and test data.
Problem train = Problem.Read(string.Format("ExtractedFeaturesTrain{0}.{1}", (a + 1).ToString(), "txt"));
Problem test = Problem.Read(string.Format("ExtractedFeaturesTest{0}.{1}", (a + 1).ToString(), "txt"));
//scalling the data
GaussianTransform gt = GaussianTransform.Compute(train);
Problem trainScaledProblem = gt.Scale(train);
Problem testScaledProblem = gt.Scale(test);
/**
* //count total number of positives and negative instances
* int trp = trainScaledProblem.Y.Count(s => s == 1);
* int trN = trainScaledProblem.Y.Count(s => s == -1);
* int tep = testScaledProblem.Y.Count(s => s == 1);
* int teN = testScaledProblem.Y.Count(s => s == -1);
* int totp = trp + tep;
* int totN = trN + teN;
**/
//For this example (and indeed, many scenarios), the default parameters will suffice.
Parameter parameters = new Parameter();
//double C = new double();
//double Gamma = new double();
Console.WriteLine("\nClassification Number {0} Step: {1}...............................\n", a + 1, r + 1);
//This will do a grid optimization to find the best parameters and store them in C and Gamma, outputting the entire
//search to params.txt.
/*
* if (a == 0)
* {
* ParameterSelection.Grid(trainScaledProblem, parameters, "params.txt", out C, out Gamma);
* CV1.Add(ParameterSelection.CVAccuracy);
* CV1.Add(C);
* CV1.Add(Gamma);
* }
* else if (a == 1)
* {
* ParameterSelection.Grid(trainScaledProblem, parameters, "params.txt", out C, out Gamma);
* CV2.Add(ParameterSelection.CVAccuracy);
* CV2.Add(C);
* CV2.Add(Gamma);
*
* if (CV1[0] > CV2[0]) //if the previous CV rate is greater than the present, then, discard the present and use the C and Gamma of the previous.
* {
* C = CV1[1];
* Gamma = CV1[2];
* }
*
* }*/
/**
* //Standard SVM (i.e. SVM without instance selection)
* ParameterSelection.Grid(trainScaledProblem, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(trainScaledProblem, parameters);
**/
//Bootstrap Sampling Method
//Training.samplingGellingPoint(trainScaledProblem, testScaledProblem);
//int subsetNumber = 5;
//int samplesPerSubset = 30;
//Problem subsets = new Problem();
//Parameter bestPara = new Parameter();
//subsets = Training.BootstrapSampling(trainScaledProblem, parameters, subsetNumber, samplesPerSubset, testScaledProblem, out bestPara); //select subsets using boothtrap sampling method
//parameters.C = C;
//parameters.Gamma = Gamma;
/**
* //KNN-Based boundary instance Selection
* KNNInstanceSelectionAlgorithm knn = new KNNInstanceSelectionAlgorithm();
* int k = 50;
* int numberOfSubset = 300; //subset to select for training
* Problem dataSubset = knn.computeNearestNeighbour(k, trainScaledProblem, numberOfSubset);
* //Problem dataSubset = knn.computeNearestNeighbour(k, trainScaledProblem, numberOfSubset);
* ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(dataSubset, parameters);
**/
/**
* AntColonyOptimizationBoundarySelection aco = new AntColonyOptimizationBoundarySelection();
* int numberOfSubset = 500; //subset to select for training
* int kNum = 300;
* //Problem dataSubset = aco.ACOBoundarySelection(train, numberOfSubset, kNum);
* Problem dataSubset = aco.ACOBoundarySelection(trainScaledProblem, numberOfSubset, kNum);
* ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(dataSubset, parameters);
**/
/**
* //FFA_Based Instance Selection
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(trainScaledProblem, out storagePercentage);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
*/
/**
* //Flower Pollination Algorithm (FPA) Based Instance Selection
* FlowerPollinationAlgorithm fpa = new FlowerPollinationAlgorithm();
* //Problem subP = fpa.FlowerPollination(trainScaledProblem);
* Problem subP = fpa.BinaryFlowerPollination(trainScaledProblem, out storagePercentage);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
**/
/**
* //Cuckoo Search Algorithm
* CuckooSearchAlgorithm csa = new CuckooSearchAlgorithm();
* Problem subP = csa.CuckooSearch(trainScaledProblem, out storagePercentage);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
**/
//Social Spider Algorithms
SocialSpiderAlgorithm ss = new SocialSpiderAlgorithm();
Problem subP = ss.SocialSpider(trainScaledProblem, out storagePercentage);
ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
parameters.C = C;
parameters.Gamma = Gamma;
Model model = Training.Train(subP, parameters);
/**
* //Bat Algorithm (BA) Based Instance Selection
* BatAlgorithm bat = new BatAlgorithm();
* Problem subP = bat.BinaryBat(trainScaledProblem, out storagePercentage);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
*/
/**
* //Clustering-Based Instance Selection Algorithm
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* ParameterSelection.Grid(boundaryInstance, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(boundaryInstance, parameters);
**/
/**
* //Edge Instance Selection
* Problem edgeNN = Training.EdgeInstanceSelection(trainScaledProblem);
* ParameterSelection.Grid(edgeNN, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(edgeNN, parameters);
**/
/**
* //Hybrid: Clustering + FFA + EISA
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(boundaryInstance);
* Problem edgeNN = Training.EdgeInstanceSelection(subP);
* ParameterSelection.Grid(edgeNN, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(edgeNN, parameters);
*/
/**
* //Hybrid: FFA + EISA
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(trainScaledProblem);
* Problem edgeNN = Training.EdgeInstanceSelection(subP);
* ParameterSelection.Grid(edgeNN, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(edgeNN, parameters);
**/
/**
* //Hybrid: KNN-based based + FFA-Based
* //Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(trainScaledProblem);
* int k = 50;
* int numberOfSubset = 100; //subset to select for training
* Problem dataSubset = Training.computeNearestNeighbour(k, subP, numberOfSubset);
* ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
*/
/**
* //Hybrid: Clustering-Based + FFA-Based
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(boundaryInstance);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(boundaryInstance, parameters);
**/
/**
* //Hybrid: Clustering-Based + Flower Pollination Algorithm
* FlowerPollinationAlgorithm fpa = new FlowerPollinationAlgorithm();
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* Problem subP = fpa.FlowerPollination(trainScaledProblem);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(boundaryInstance, parameters);
**/
/* //Hybrid: Clustering based + FFA-Based + KNN-Based
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(boundaryInstance);
* int k = 50;
* int numberOfSubset = 100; //subset to select for training
* Problem dataSubset = Training.computeNearestNeighbour(k, boundaryInstance, numberOfSubset);
* ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(dataSubset, parameters);
*/
//Train the model using the optimal parameters.
//Model model = Training.Train(trainScaledProblem, parameters);
//removing support vectors that contributes less to the decision surface
//Model submod = Training.performSupportVectorReduction(model, trainScaledProblem);
//Perform classification on the test data, putting the results in results.txt.
//classficatnAccuracySum += Prediction.Predict(testScaledProblem, "ClassificationResults.txt", model, false);
//classficatnAccuracy = Prediction.Predict(test, "ClassificationResults.txt", model, false); //classfication accuracy for each iteration ->for the purpose of outputting to the text file
classficatnAccuracy = Prediction.Predict(testScaledProblem, "ClassificationResults.txt", model, false); //classfication accuracy for each iteration ->for the purpose of outputting to the text file
classficatnAccuracySum += classficatnAccuracy;
Console.WriteLine("\nClassification Accuracy: {0}%", 100 * classficatnAccuracy);
PerformanceEvaluator pp = new PerformanceEvaluator("ClassificationResults.txt", test, out TP, out TN, out FP, out FN, out P, out F_M);
DateTime end1 = DateTime.Now;
TimeSpan duration1 = end1 - start1;
double time1 = duration1.Minutes * 60.0 + duration1.Seconds + duration1.Milliseconds / 1000.0;
sumTP += TP; sumTN += TN; sumFP += FP; sumFN += FN; sumP += P; sumF_M += F_M; sumTime += time1; sumStoragePercentage += storagePercentage;
//saving all the output to a file
string outpt = string.Format("Cross Validation: {0}, Run number {1}, CAccuracy: {2:0.0000} FP: {3:0.0000}, FN: {4:0.0000}, Recall: {5:0.0000}, Precision: {6:0.0000}, FMeasure: {7:0.0000}, Time: {8} Seconds, Storage Percentage: {9}", a + 1, r + 1, (classficatnAccuracy * 100), (FP * 100), (FN * 100), (TP * 100), (P * 100), (F_M * 100), time1, storagePercentage);
File.AppendAllText(outputEvaluationFilePath, outpt);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
}
if (classficatnAccuracy * 100 > globalBest)
{
globalBest = classficatnAccuracy * 100;
}
}
classficatnAccuracySum = (classficatnAccuracySum * 100) / N_Fold; //converting to percentage and dividing by the number of folds
sumFP = (sumFP * 100) / N_Fold; //calculating the average cross validation for False Positive over 10 folds
sumTP = (sumTP * 100) / N_Fold; //calculating the average cross validation for Recall over 10 folds
sumFN = (sumFN * 100) / N_Fold; //calculating the average cross validation for False Negative over 10 folds
sumF_M = (sumF_M * 100) / N_Fold; //calculating the average cross validation for F Measure over 10 folds
sumP = (sumP * 100) / N_Fold; //calculating the average cross validation for Precision over 10 folds
sumStoragePercentage = sumStoragePercentage / N_Fold;
avgRuns += classficatnAccuracySum;
avgFP += sumFP;
avgFN += sumFN;
avgR += sumTP;
avgPr += sumP;
avgFM += sumF_M;
avgTime += sumTime;
avgStoragePercentage += sumStoragePercentage;
//saving all the outputs to a file
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, string.Format("Average Calculations....Run Number: {0}", r + 1));
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
string outpt2 = string.Format("Run number {0}, Average CAccuracy: {1:0.0000} FP: {2:0.0000}, FN: {3:0.0000}, Recall: {4:0.0000}, Precision: {5:0.0000}, FMeasure: {6:0.0000}, Time: {7}, Storage Percentage: {8}", r + 1, classficatnAccuracySum, sumFP, sumFN, sumTP, sumP, sumF_M, sumTime, sumStoragePercentage);
File.AppendAllText(outputEvaluationFilePath, outpt2);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
Console.WriteLine("\nStep {0}...............................\n", r + 1);
}
}
//}
DateTime end = DateTime.Now;
TimeSpan duration = end - start;
double time = duration.Minutes * 60.0 + duration.Seconds + duration.Milliseconds / 1000.0;
Console.WriteLine("\nAverage processing time {0:########.00} seconds\n", avgTime / n_Runs);
//Console.WriteLine("\nTotal processing time {0:########.00} seconds\n", time);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, "Average processing time:\n" + avgTime / n_Runs + " Seconds");
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
//sending all the outputs to the screen
Console.WriteLine("\nOverall Average Accuracy: {0:0.00}% \nGlobal Best: {1:0.00}%", avgRuns / n_Runs, globalBest);
Console.WriteLine("\n\nTotal False Positive: {0:0.00}%\nTotal False Negative: {1:0.00}%\nRecall: {2:0.00}%\nPrecision: {3:0.00}%\nF_Measure: {4:0.00}% \nStorage Percentage: {5:0.00}%", (avgFP / n_Runs), (avgFN / n_Runs), (avgR / n_Runs), (avgPr / n_Runs), (avgFM / n_Runs), (avgStoragePercentage / n_Runs));
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, "Overall Average Calculations.......");
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
string outpt3 = string.Format("Overall Average CAccuracy: {0:0.0000} FP: {1:0.0000}, FN: {2:0.0000}, Recall: {3:0.0000}, Precision: {4:0.0000}, FMeasure: {5:0.0000}, Storage Percentage: {6:0.0000}", avgRuns / n_Runs, avgFP / n_Runs, avgFN / n_Runs, avgR / n_Runs, avgPr / n_Runs, avgFM / n_Runs, avgStoragePercentage / n_Runs);
File.AppendAllText(outputEvaluationFilePath, outpt3);
Console.ReadKey();
}
static void Main(string[] args)
{
PDSClass pds = new PDSClass();
FireFlyAlgorithm ff = new FireFlyAlgorithm();
ParameterSelection ps = new ParameterSelection();
//reading the phishing emails and the test Datas
string testFileFolderName = "TrainTestData";
string trainTestDataPath = String.Format(Environment.CurrentDirectory + "\\{0}", testFileFolderName); //filepath for the training and test dataset
string[] trainTestFileURL = System.IO.Directory.GetFileSystemEntries(trainTestDataPath);
string extractedVectorsFolderName = "ExtractedFeatures";
string extractedVectorsFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", extractedVectorsFolderName);
string[] evFileURL = System.IO.Directory.GetFileSystemEntries(extractedVectorsFilePath);
string outputEvaluationFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "OutputEvaluations.txt");
//string SVAccuracyFilePath = String.Format(Environment.CurrentDirectory + "\\{0}", "SVAccuracy.txt");
File.WriteAllText(ps.SVAccuracyFilePath, string.Empty);
File.WriteAllText(outputEvaluationFilePath, string.Empty); //deleting the contents of the file holding the results for new results
File.WriteAllText(ps.filePath, string.Empty); //deleting the contents of the file holding the extracted C,Gamma and CV values
File.WriteAllText(pds.extractedFeaturesFilePathTrainDS, string.Empty);
File.WriteAllText(ps.filePath2, string.Empty);
File.WriteAllText(ps.filePath3, string.Empty);
int NumofFeatures = 17;
double DatasetEntropy = 0.0;
int TotalMails;
string[] features = pds.Features(); //saving the features in the string array
int[,] HamPhishCount = new int[, ] {
}; //count for the total number of ham and phish emails
string[] testTrainFolderFiles = new string[] { };
string folderFiles;
double[] informationGain = new double[NumofFeatures];
Dictionary <string, double> feat_infoGain = pds.Feature_InfoGain(features, informationGain); //saving the information gain for all the individual features
double classficatnAccuracySum = 0.0; //cummulative classification accuracy
double classficatnAccuracy = 0; //classification accuracy for each iteration
DateTime start = DateTime.Now;
double totalPhishing = 0, totalHam = 0; int mailGrandTotal = 0;
//initializing all the variables that will used for evaluating the performance of the classifier
double FP = 0.0, FN = 0.0, TP = 0.0, TN = 0.0, P = 0.0, F_M = 0.0, sumFP = 0.0, sumF_M = 0.0, sumFN = 0.0, sumTP = 0.0, sumTN = 0.0, sumP = 0.0;
int N_Fold = 10; //number of folds
int n_Runs = 1; //number of runs
double[,] NormalizedVector = new double[, ] {
}; //normalized vector values for each features
Program p = new Program();
double avgRuns = 0.0, avgFP = 0.0, avgFN = 0.0, avgR = 0.0, avgPr = 0.0, avgFM = 0.0; //avg=> average
double globalBest = double.MinValue;
//change the boolean value appropriately to choose the task you want to perform (either vector value extraction or email classification)
//both values should not be true. This is to reduce processing time
bool extractVectorValues = true;
bool Emailclassification = false;
double C = new double();
double Gamma = new double();
//double[] CV = new double[2];
List <double> CV1 = new List <double>(); //Save the CV accuracy, C and Gamma for comparison
List <double> CV2 = new List <double>(); //Save the CV accuracy, C and Gamma for comparison
for (int aa = 0; aa < n_Runs; aa++)
{
classficatnAccuracySum = 0.0;
sumFP = 0.0;
sumTP = 0.0; //Recall
sumFN = 0.0;
sumF_M = 0.0;
sumP = 0.0;
for (int a = 0; a < N_Fold; a++)
{
if (extractVectorValues == true) //if the value of ExtractVectorValues is true, only extract email vector values and dont perform classification
{
n_Runs = 1; // change number of runs from its default value (i.e 10) to 1 (to avoid repeating the extraction process 10 times) since we wanna do is to extract the vector values from each emails
string[] trainFileURLs = new string[] { }; //urls for the train emails (i.e. the emails used for training the classifier)
string[] testFileURLs = new string[] { }; //urls for the test emails (i.e. the emails used for testing the classifier)
Dictionary <string, int> trainMail_Class = new Dictionary <string, int>(); //variable containing the emails and classes of all the training emails
Dictionary <string, int> testMail_Class = new Dictionary <string, int>(); //variable containing the emails and classes of all the test emails
string[] trainMailFileNames = new string[trainMail_Class.Count]; //the file names for all the emails in the training dataset
string[] testMailFileNames = new string[] { }; //the file names for all the emails in the test dataset
string trainMailLabel; int phishCount = 0, hamCount = 0; double phishPercentage, hamPercentage;
//processing the training dataset for the each fold
for (int i = 0; i < trainTestFileURL.Length; i++)
{
if (i.Equals(a))
{
continue; //skipping one email folder, which is to be used for testing the trained classifier (i.e. the current test dataset)
}
testTrainFolderFiles = System.IO.Directory.GetFiles(trainTestFileURL[i]); //getting the filenames of all the emails in the training dataset
trainFileURLs = trainFileURLs.Concat(testTrainFolderFiles).ToArray(); //get all the urls for the test emails
trainMailFileNames = pds.getFileNames(trainFileURLs); //get the file names for all the test mails
for (int j = 0; j < testTrainFolderFiles.Length; j++) //processing all the emails in the current training dataset for classification
{
trainMailLabel = trainMailFileNames[j].Substring(0, 2); //getting the label for each email, HM(for Ham Mails) and PM(for Phish Mails)
folderFiles = File.ReadAllText(testTrainFolderFiles[j]); //extracting the content of each email in each email folder
trainMail_Class[pds.ProcessMails(folderFiles)] = (trainMailLabel.Equals("PM")) ? 1 : 0; //processing each email and assigning label to the emails based on the folders each emails come from.
if (trainMail_Class.ElementAt(j).Value == 1)
{
phishCount++; //counting the total number of ham and phishing to get their percentage
}
else
{
hamCount++;
}
}
}
//processing the test dataset for each fold
for (int i = a; i < a + 1; i++)
{
testTrainFolderFiles = System.IO.Directory.GetFiles(trainTestFileURL[i]);
testFileURLs = testFileURLs.Concat(testTrainFolderFiles).ToArray();
testMailFileNames = pds.getFileNames(testFileURLs);
for (int j = 0; j < testTrainFolderFiles.Length; j++)
{
trainMailLabel = testMailFileNames[j].Substring(0, 2);
folderFiles = File.ReadAllText(testTrainFolderFiles[j]);
testMail_Class[pds.ProcessMails(folderFiles)] = (trainMailLabel.Equals("PM")) ? 1 : 0; //processing each email and assigning label to the emails based on the folders each emails come from.
if (testMail_Class.ElementAt(j).Value == 1)
{
phishCount++;
}
else
{
hamCount++;
}
}
}
//calculating the percentage of phishing and ham email in the dataset
phishPercentage = (double)phishCount / (double)(trainMail_Class.Count + testMail_Class.Count);
hamPercentage = (double)hamCount / (double)(trainMail_Class.Count + testMail_Class.Count);
mailGrandTotal = phishCount + hamCount;
totalHam = hamCount; totalPhishing = phishCount;
//Information Gain
TotalMails = trainMail_Class.Count;
int[,] vector = new int[TotalMails, NumofFeatures];
pds.processVector(vector, trainMail_Class, features, trainFileURLs, NumofFeatures); //extracting the vector values of all the features
HamPhishCount = new int[NumofFeatures, 4];
pds.FeatureVectorSum(NumofFeatures, TotalMails, vector, trainMail_Class, HamPhishCount); // calculating the total number of zeros and ones for both phishing and ham emails
DatasetEntropy = pds.Entropy(trainMail_Class); //calculating the entropy for the entire dataset
pds.CalInformationGain(NumofFeatures, HamPhishCount, informationGain, TotalMails, DatasetEntropy); //calculating information gain for each feature
feat_infoGain = pds.Feature_InfoGain(features, informationGain); //assisgning the calculated information gain to each feature
//process vector for training Dataset
int NumofFeatures2 = NumofFeatures - 8;
string[] newFeatures = new string[NumofFeatures2];
for (int i = 0; i < NumofFeatures2; i++)
{
newFeatures[i] = feat_infoGain.ElementAt(i).Key; //copying the best 8 features with the highest information gain
}
TotalMails = trainMail_Class.Count;
vector = new int[TotalMails, NumofFeatures2];
pds.processVector(vector, trainMail_Class, newFeatures, trainFileURLs, NumofFeatures2);
NormalizedVector = ff.Normalize(vector); //normalize the all vector values for train data
//extract vectors of the training data
pds.extractVectors(vector, trainMail_Class, NumofFeatures2, "trainingDS", a);
//process vector for testing Dataset
TotalMails = testMail_Class.Count;
vector = new int[TotalMails, NumofFeatures2];
pds.processVector(vector, testMail_Class, newFeatures, testFileURLs, NumofFeatures2);
NormalizedVector = ff.Normalize(vector); //normalize the all vector values for test data
//extract vectors of the test data
pds.extractVectors(vector, testMail_Class, NumofFeatures2, "testDS", a);
}
if (Emailclassification == true) //if the value of EmailClassification is true, perform email classification and dont extract emails
{
//SVM Classfication begins here
//First, read in the training and test data.
Problem train = Problem.Read(string.Format("ExtractedFeaturesTrain{0}.{1}", (a + 1).ToString(), "txt"));
Problem test = Problem.Read(string.Format("ExtractedFeaturesTest{0}.{1}", (a + 1).ToString(), "txt"));
//scalling the data
GaussianTransform gt = GaussianTransform.Compute(train);
Problem trainScaledProblem = gt.Scale(train);
Problem testScaledProblem = gt.Scale(test);
//For this example (and indeed, many scenarios), the default parameters will suffice.
Parameter parameters = new Parameter();
//double C = new double();
//double Gamma = new double();
Console.WriteLine("\nClassification Number {0} Step: {1}...............................\n", a + 1, aa + 1);
//This will do a grid optimization to find the best parameters and store them in C and Gamma, outputting the entire
//search to params.txt.
/*
* if (a == 0)
* {
* ParameterSelection.Grid(trainScaledProblem, parameters, "params.txt", out C, out Gamma);
* CV1.Add(ParameterSelection.CVAccuracy);
* CV1.Add(C);
* CV1.Add(Gamma);
* }
* else if (a == 1)
* {
* ParameterSelection.Grid(trainScaledProblem, parameters, "params.txt", out C, out Gamma);
* CV2.Add(ParameterSelection.CVAccuracy);
* CV2.Add(C);
* CV2.Add(Gamma);
*
* if (CV1[0] > CV2[0]) //if the previous CV rate is greater than the present, then, discard the present and use the C and Gamma of the previous.
* {
* C = CV1[1];
* Gamma = CV1[2];
* }
*
* }*/
//Bootstrap Sampling Method
//Training.samplingGellingPoint(trainScaledProblem, testScaledProblem);
//int subsetNumber = 5;
//int samplesPerSubset = 30;
//Problem subsets = new Problem();
//Parameter bestPara = new Parameter();
//subsets = Training.BootstrapSampling(trainScaledProblem, parameters, subsetNumber, samplesPerSubset, testScaledProblem, out bestPara); //select subsets using boothtrap sampling method
//parameters.C = C;
//parameters.Gamma = Gamma;
//KNN-Based boundary instance Selection
int k = 50;
int numberOfSubset = 300; //subset to select for training
Problem dataSubset = Training.computeNearestNeighbour(k, trainScaledProblem, numberOfSubset);
ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
parameters.C = C;
parameters.Gamma = Gamma;
//Model model = Training.buildModel(dataSubset, parameters);
Model model = Training.Train(dataSubset, parameters);
//ParameterSelection.Grid(boundaryInstance, parameters, "params.txt", out C, out Gamma);
//ParameterSelection.Grid(trainScaledProblem, parameters, "params.txt", out C, out Gamma);
/* //FFA_Based Instance Selection
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(trainScaledProblem);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
*/
/*
* //Clustering-Based Instance Selection Algorithm
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* ParameterSelection.Grid(boundaryInstance, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(boundaryInstance, parameters);
*/
/* //Edge Instance Selection
* Problem edgeNN = Training.EdgeInstanceSelection(trainScaledProblem);
* ParameterSelection.Grid(edgeNN, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(edgeNN, parameters);
*/
/*
* //Hybrid: KNN-based based + FFA-Based
* //Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(trainScaledProblem);
* int k = 50;
* int numberOfSubset = 100; //subset to select for training
* Problem dataSubset = Training.computeNearestNeighbour(k, subP, numberOfSubset);
* ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(subP, parameters);
*/
/*
* //Hybrid: Clustering-Based + FFA-Based
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(boundaryInstance);
* ParameterSelection.Grid(subP, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(boundaryInstance, parameters);
*/
/* //Hybrid: Clustering based + FFA-Based + KNN-Based
* Problem boundaryInstance = Training.ClusteringBoundaryInstance(trainScaledProblem);
* FireflyInstanceSelection fi = new FireflyInstanceSelection();
* Problem subP = fi.firefly_simple(boundaryInstance);
* int k = 50;
* int numberOfSubset = 100; //subset to select for training
* Problem dataSubset = Training.computeNearestNeighbour(k, boundaryInstance, numberOfSubset);
* ParameterSelection.Grid(dataSubset, parameters, "params.txt", out C, out Gamma);
* parameters.C = C;
* parameters.Gamma = Gamma;
* Model model = Training.Train(dataSubset, parameters);
*/
//Train the model using the optimal parameters.
//Model model = Training.Train(trainScaledProblem, parameters);
//removing support vectors that contributes less to the decision surface
//Model submod = Training.performSupportVectorReduction(model, trainScaledProblem);
//Perform classification on the test data, putting the results in results.txt.
//classficatnAccuracySum += Prediction.Predict(testScaledProblem, "ClassificationResults.txt", model, false);
classficatnAccuracy = Prediction.Predict(testScaledProblem, "ClassificationResults.txt", model, false); //classfication accuracy for each iteration ->for the purpose of outputting to the text file
classficatnAccuracySum += classficatnAccuracy;
Console.WriteLine("\nClassification Accuracy: {0}%", 100 * classficatnAccuracy);
PerformanceEvaluator pp = new PerformanceEvaluator("ClassificationResults.txt", test, out TP, out TN, out FP, out FN, out P, out F_M);
sumTP += TP; sumTN += TN; sumFP += FP; sumFN += FN; sumP += P; sumF_M += F_M;
//saving all the output to a file
string outpt = string.Format("Cross Validation: {0}, Run number {1}, CAccuracy: {2:0.0000} FP: {3:0.0000}, FN: {4:0.0000}, Recall: {5:0.0000}, Precision: {6:0.0000}, FMeasure: {7:0.0000}", a + 1, aa + 1, (classficatnAccuracy * 100), (FP * 100), (FN * 100), (TP * 100), (P * 100), (F_M * 100));
File.AppendAllText(outputEvaluationFilePath, outpt);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
}
if (classficatnAccuracy * 100 > globalBest)
{
globalBest = classficatnAccuracy * 100;
}
}
classficatnAccuracySum = (classficatnAccuracySum * 100) / N_Fold; //converting to percentage and dividing by the number of folds
sumFP = (sumFP * 100) / N_Fold; //calculating the average cross validation for False Positive over 10 folds
sumTP = (sumTP * 100) / N_Fold; //calculating the average cross validation for Recall over 10 folds
sumFN = (sumFN * 100) / N_Fold; //calculating the average cross validation for False Negative over 10 folds
sumF_M = (sumF_M * 100) / N_Fold; //calculating the average cross validation for F Measure over 10 folds
sumP = (sumP * 100) / N_Fold; //calculating the average cross validation for Precision over 10 folds
avgRuns += classficatnAccuracySum;
avgFP += sumFP;
avgFN += sumFN;
avgR += sumTP;
avgPr += sumP;
avgFM += sumF_M;
//saving all the outputs to a file
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, string.Format("Average Calculations....Run Number: {0}", aa + 1));
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
string outpt2 = string.Format("Run number {0}, Average CAccuracy: {1:0.0000} FP: {2:0.0000}, FN: {3:0.0000}, Recall: {4:0.0000}, Precision: {5:0.0000}, FMeasure: {6:0.0000}", aa + 1, classficatnAccuracySum, sumFP, sumFN, sumTP, sumP, sumF_M);
File.AppendAllText(outputEvaluationFilePath, outpt2);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
Console.WriteLine("\nStep {0}...............................\n", aa + 1);
}
DateTime end = DateTime.Now;
TimeSpan duration = end - start;
double time = duration.Minutes * 60.0 + duration.Seconds + duration.Milliseconds / 1000.0;
Console.WriteLine("\nTotal processing time {0:########.00} seconds\n", time);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, "Total processing time:\n" + time + " Seconds");
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
//sending all the outputs to the screen
Console.WriteLine("\nOverall Average Accuracy: {0:0.00}% \nGlobal Best: {1:0.00}%", avgRuns / n_Runs, globalBest);
Console.WriteLine("\n\nTotal False Positive: {0:0.00}%\nTotal False Negative: {1:0.00}%\nRecall: {2:0.00}%\nPrecision: {3:0.00}%\nF_Measure: {4:0.00}%", (avgFP / n_Runs), (avgFN / n_Runs), (avgR / n_Runs), (avgPr / n_Runs), (avgFM / n_Runs));
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, "Overall Average Calculations.......");
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
File.AppendAllText(outputEvaluationFilePath, Environment.NewLine);
string outpt3 = string.Format("Overall Average CAccuracy: {0:0.0000} FP: {1:0.0000}, FN: {2:0.0000}, Recall: {3:0.0000}, Precision: {4:0.0000}, FMeasure: {5:0.0000}", avgRuns / n_Runs, avgFP / n_Runs, avgFN / n_Runs, avgR / n_Runs, avgPr / n_Runs, avgFM / n_Runs);
File.AppendAllText(outputEvaluationFilePath, outpt3);
Console.ReadKey();
}