public static void Run_Classify()
{
List <CDataRecord> records = LoadSample();
DataTransformer <CDataRecord> dt = new DataTransformer <CDataRecord>();
dt.DoFeaturesScaling(records);
KernelSVM <CDataRecord> algorithm = new KernelSVM <CDataRecord>();
algorithm.C = 20;
((GaussianKernel)algorithm.Kernel).Sigma = 0.01;
algorithm.Train(records);
Console.WriteLine("SVM(Gaussian Kernel) Model Built!");
for (int i = 0; i < records.Count; i++)
{
CDataRecord rec = records[i] as CDataRecord;
Console.WriteLine("rec: ");
for (int j = 0; j < rec.Dimension; ++j)
{
Console.WriteLine("X[" + j + "] = " + "[" + rec[j] + "] ");
}
Console.WriteLine("Label: " + "[" + rec.Label + "] ");
Console.WriteLine("Predicted Label: " + algorithm.Predict(records[i]));
}
}
public static void Run_Rank()
{
List <CDataRecord> records = LoadSample();
DataTransformer <CDataRecord> dt = new DataTransformer <CDataRecord>();
dt.DoFeaturesScaling(records);
KernelSVM <CDataRecord> algorithm = new KernelSVM <CDataRecord>();
algorithm.C = 20; //large value, high bias
((GaussianKernel)algorithm.Kernel).Sigma = 0.01; //low value, high bias
algorithm.Train(records);
Console.WriteLine("SVM (Gaussian Kernel) Model Built!");
for (int i = 0; i < records.Count; i++)
{
CDataRecord rec = records[i] as CDataRecord;
Console.WriteLine("rec: ");
for (int j = 0; j < rec.Dimension; ++j)
{
Console.WriteLine("X[" + j + " = " + rec[j] + "]");
}
Console.WriteLine("Label: " + rec.Label);
List <KeyValuePair <string, double> > ranks = algorithm.Rank(records, rec);
for (int k = 0; k < ranks.Count; ++k)
{
Console.WriteLine("{0}: score = {1}", ranks[k].Key, ranks[k].Value);
}
}
}
private void ImportData(string X_filepath, string y_filepath)
{
mTrainingPercent = 100;
mTestingPercent = 0;
mCrossValidationPercent = 0;
lvTabularData.Items.Clear();
lvTabularData.Columns.Clear();
List <MLDataPoint> X_points = MLDataPointUtil.LoadDataSet(X_filepath);
List <List <int> > y_points = IntDataTableUtil.LoadDataSet(y_filepath);
List <CDataRecord> X = new List <CDataRecord>();
for (int i = 0; i < X_points.Count; ++i)
{
CDataRecord X_i = new CDataRecord(X_points[i].Dimension);
for (int j = 0; j < X_points[i].Dimension; ++j)
{
X_i[j + 1] = X_points[i][j];
}
X_i.Label = y_points[i][0].ToString();
X.Add(X_i);
}
lvTabularData.Columns.Add("#");
for (int i = 0; i < X[0].Dimension; ++i)
{
lvTabularData.Columns.Add(string.Format("X[{0}]", i));
}
lvTabularData.Columns.Add("Y");
lvTabularData.Columns.Add("Data Set");
lvTabularData.Columns.Add("Predicted Y");
int m = X.Count;
for (int i = 0; i < m; ++i)
{
CDataRecord X_i = X[i];
ListViewItem item = new ListViewItem();
item.Text = (lvTabularData.Items.Count + 1).ToString();
for (int j = 0; j < X_i.Dimension; ++j)
{
item.SubItems.Add(X_i[j].ToString());
}
item.SubItems.Add(X_i.Label.ToString());
item.SubItems.Add(X_i.DataSetType.ToString());
item.SubItems.Add(X_i.PredictedLabel.ToString());
item.Tag = X_i;
item.ForeColor = Color.Green;
lvTabularData.Items.Add(item);
}
}
private void btnSplitData_Click(object sender, EventArgs e)
{
FrmDataSplitOption dlg = new FrmDataSplitOption();
dlg.TrainingPercent = mTrainingPercent;
dlg.TestingPercent = mTestingPercent;
dlg.CrossValidationPercent = mCrossValidationPercent;
if (dlg.ShowDialog() == DialogResult.OK)
{
double training_percent = dlg.TrainingPercent;
double testing_percent = dlg.TestingPercent;
double crossvalidation_percent = dlg.CrossValidationPercent;
double sum = training_percent + testing_percent + crossvalidation_percent;
if (sum == 0)
{
mTrainingPercent = 100;
mTestingPercent = 0;
mCrossValidationPercent = 0;
}
else
{
mTrainingPercent = training_percent * 100 / sum;
mTestingPercent = testing_percent * 100 / sum;
mCrossValidationPercent = crossvalidation_percent * 100 / sum;
}
int item_count = lvTabularData.Items.Count;
int training_pont = (int)(item_count * mTrainingPercent / 100);
int testing_point = (int)(item_count * mTestingPercent / 100) + training_pont;
for (int item_index = 0; item_index < item_count; ++item_index)
{
ListViewItem item = lvTabularData.Items[item_index];
CDataRecord rec = item.Tag as CDataRecord;
if (item_index < training_pont)
{
rec.DataSetType = DataSetTypes.Training;
}
else if (item_index < testing_point)
{
rec.DataSetType = DataSetTypes.Testing;
}
else
{
rec.DataSetType = DataSetTypes.CrossValidation;
}
}
UpdateTabularDataView();
}
}
public void Shuffle(List <CDataRecord> list)
{
Random rng = new Random();
int n = list.Count;
while (n > 1)
{
n--;
int k = rng.Next(n + 1);
CDataRecord value = list[k];
list[k] = list[n];
list[n] = value;
}
}
private List <CDataRecord> LoadDataSet(DataSetTypes dstype)
{
List <CDataRecord> data_set = new List <CDataRecord>();
foreach (ListViewItem item in lvTabularData.Items)
{
CDataRecord rec = item.Tag as CDataRecord;
if (rec.DataSetType == dstype)
{
data_set.Add(rec);
}
}
return(data_set);
}
private void btnShuffleData_Click(object sender, EventArgs e)
{
if (lvTabularData.Items.Count == 0)
{
return;
}
List <CDataRecord> data_set = new List <CDataRecord>();
foreach (ListViewItem item in lvTabularData.Items)
{
data_set.Add(item.Tag as CDataRecord);
}
Shuffle(data_set);
int item_count = lvTabularData.Items.Count;
for (int k = 0; k < item_count; ++k)
{
ListViewItem item = lvTabularData.Items[k];
CDataRecord rec = data_set[k];
item.Text = (k + 1).ToString();
int col_index = 1;
for (col_index = 1; col_index < rec.Dimension; ++col_index)
{
item.SubItems[col_index].Text = (rec[col_index].ToString());
}
item.SubItems[col_index++].Text = rec.Label.ToString();
item.SubItems[col_index++].Text = rec.DataSetType.ToString();
item.SubItems[col_index++].Text = rec.PredictedLabel.ToString();
if (rec.DataSetType == DataSetTypes.Training)
{
item.ForeColor = Color.Green;
}
else if (rec.DataSetType == DataSetTypes.Testing)
{
item.ForeColor = Color.Red;
}
else
{
item.ForeColor = Color.Blue;
}
item.Tag = rec;
}
}
public static List <CDataRecord> LoadSample()
{
XmlDocument doc = new XmlDocument();
doc.Load("database.xml");
List <CDataRecord> records = new List <CDataRecord>();
XmlElement xml_root = doc.DocumentElement;
foreach (XmlElement xml_level1 in xml_root.ChildNodes)
{
if (xml_level1.Name == "record")
{
string outlook_text = xml_level1.Attributes["outlook"].Value;
double outlook = DistributionModel.GetUniform() * 0.333;
if (outlook_text == "Overcast")
{
outlook = 0.333 + DistributionModel.GetUniform() * 0.333;
}
else if (outlook_text == "Rain")
{
outlook = 0.666 + DistributionModel.GetUniform() * 0.333;
}
double temperature = double.Parse(xml_level1.Attributes["temperature"].Value);
double humidity = double.Parse(xml_level1.Attributes["humidity"].Value);
string windy_text = xml_level1.Attributes["windy"].Value;
double windy = windy_text == "true" ? 1 : 0;
String class_label = xml_level1.Attributes["class"].Value;
CDataRecord rec = new CDataRecord(4);
//index must start at 1
rec[1] = outlook;
rec[2] = windy;
rec[3] = temperature;
rec[4] = humidity;
rec.Label = class_label;
records.Add(rec);
}
}
return(records);
}
public override double ComputeCost(List <T> data_set)
{
int sample_count = data_set.Count;
if (sample_count == 0)
{
return(-1);
}
int dimension = data_set[0].Dimension;
double[,] X = new double[sample_count, dimension];
int[] Y = new int[sample_count];
for (int i = 0; i < sample_count; ++i)
{
T rec = data_set[i];
for (int d = 0; d < dimension; ++d)
{
X[i, d] = rec[d];
}
}
double total_error = 0;
foreach (string class_label in mClassFieldLabels)
{
for (int i = 0; i < sample_count; ++i)
{
CDataRecord rec = data_set[i] as CDataRecord;
Y[i] = rec.Label == class_label ? 1 : 0;
}
LinearSVMCostFunction f = new LinearSVMCostFunction(X, Y, dimension, sample_count);
f.C = mC;
double error = f.Evaluate(mTheta[class_label]);
total_error += error;
}
return(total_error);
}
private void UpdateTabularDataView()
{
int item_count = lvTabularData.Items.Count;
int training_pont = (int)(item_count * mTrainingPercent / 100);
int testing_point = (int)(item_count * mTestingPercent / 100) + training_pont;
for (int item_index = 0; item_index < item_count; ++item_index)
{
ListViewItem item = lvTabularData.Items[item_index];
CDataRecord rec = item.Tag as CDataRecord;
item.Text = (item_index + 1).ToString();
int col_index = 1;
for (col_index = 0; col_index < rec.Dimension; ++col_index)
{
item.SubItems[col_index + 1].Text = (rec[col_index].ToString());
}
col_index++;
item.SubItems[col_index++].Text = rec.Label.ToString();
item.SubItems[col_index++].Text = rec.DataSetType.ToString();
item.SubItems[col_index++].Text = rec.PredictedLabel.ToString();
if (rec.DataSetType == DataSetTypes.Training)
{
item.ForeColor = Color.Green;
}
else if (rec.DataSetType == DataSetTypes.Testing)
{
item.ForeColor = Color.Red;
}
else
{
item.ForeColor = Color.Blue;
}
}
}
public static void Run_Classify2()
{
List <MLDataPoint> X_points = MLDataPointUtil.LoadDataSet("X1.txt");
List <List <int> > y_points = IntDataTableUtil.LoadDataSet("y1.txt");
List <CDataRecord> X = new List <CDataRecord>();
for (int i = 0; i < X_points.Count; ++i)
{
CDataRecord X_i = new CDataRecord(X_points[i].Dimension);
for (int j = 0; j < X_points[i].Dimension; ++j)
{
X_i[i + 1] = X_points[i][j]; //X_i index must start at 1
}
X_i.Label = y_points[i][0].ToString();
X.Add(X_i);
}
LinearSVM <CDataRecord> algorithm = new LinearSVM <CDataRecord>();
algorithm.C = 100;
algorithm.Train(X);
}
