public Perceptron(List <Entry> train, List <Entry> test, double learning_rate, bool dymanicLearningRate, double margin, WeightBias wb_average, bool aggressive,
double c, bool svm, double tradeoff, bool logistic_regression, int forestSize)
{
Training_Data = train;
Test_Data = test;
Learning_Rate = learning_rate;
Initial_Learning_Rate = learning_rate;
DymanicLearningRate = dymanicLearningRate;
if (DymanicLearningRate)
{
T_Count = 1;
}
Margin = margin;
if (wb_average != null)
{
WeightBias_Average = wb_average;
}
Aggressive = aggressive;
Labels = new List <int>();
C = c;
SVM = svm;
Tradeoff = tradeoff;
Logistic_Regression = logistic_regression;
ForestSize = forestSize;
}
public Data(List <Entry> r1, List <Entry> r2, Random r, int epochs, double learning_rate, double margin, double c, bool logistic_regression, double tradeoff)
{
C = c;
Tradeoff = tradeoff;
Training_Data = r1;
Test_Data = r2;
AccuracyWeightB = new Dictionary <int, AccuracyWB>();
perceptron = new Perceptron(Training_Data, Test_Data, learning_rate, margin, C, logistic_regression, Tradeoff, r);
Dictionary <int, double> w = new Dictionary <int, double>();
double b = (r.NextDouble() * (0.01 + 0.01) - 0.01);
WeightBias wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
AccuracyWeightB.Add(i + 1, new AccuracyWB(perceptron.GetAccuracy(Test_Data, wb), wb));
perceptron.ShuffleTraining_Data(r);
}
AccuracyWB bestAccuracy = AccuracyWeightB.OrderByDescending(x => x.Value.Accuracy).ThenByDescending(y => y.Key).Select(z => z.Value).First();
Train_Accuracy = perceptron.GetAccuracy(Training_Data, bestAccuracy.Weight_Bias); //Train Accuracy
Test_Accuracy = bestAccuracy.Accuracy; //Test Accuracy
BestWeightBias = bestAccuracy.Weight_Bias;
Learning_Rate = learning_rate;
}
public double GetAccuracy(List <Entry> test_Data, WeightBias wb)
{
double[] w = wb.Weight;
double b = wb.Bias;
double TotalErrors = 0;
foreach (var item in test_Data)
{
int y = item.Sign;
int yguess;
double[] x = item.Vector;
double xw = 0;
for (int i = 0; i < ForestSize; i++)
{
xw = xw + (x[i] * w[i]);
}
xw += b;
if (xw >= 0)
{
yguess = +1;
Labels.Add(yguess);
}
else
{
yguess = -1;
Labels.Add(0);
}
if (y != yguess)
{
TotalErrors++;
}
}
return(100 - ((TotalErrors / Convert.ToDouble(test_Data.Count)) * 100));
}
public Data(List <Entry> r, StreamReader r2, double learning_rate, WeightBias bestWB, int forestSize)
{
ForestSize = forestSize;
data_1 = r;
data_2 = new List <Entry>();
AccuracyWeightB = new Dictionary <int, AccuracyWB>();
Predictions = new List <Prediction>();
//SetData(r);
perceptron = new Perceptron(data_1, null, learning_rate, false, 0, null, false, 0, false, 0, false, ForestSize);
Test_Accuracy = perceptron.GetAccuracy(data_1, bestWB);
SetAccountIDs(r2, perceptron.Labels);
}
public double GetAccuracy(List <Entry> test_Data, WeightBias wb)
{
Dictionary <int, double> w = wb.Weight;
double b = wb.Bias;
double TotalErrors = 0;
foreach (var item in test_Data)
{
int y = item.Sign;
int yguess;
Dictionary <int, double> x = item.Vector;
double xw = 0;
foreach (var xi in x)
{
if (w.ContainsKey(xi.Key))
{
xw = xw + (w[xi.Key] * xi.Value);
}
}
xw += b;
if (xw >= 0)
{
yguess = +1;
}
else
{
yguess = -1;
}
Labels.Add(yguess);
if (y != yguess)
{
TotalErrors++;
}
}
return(100 - ((TotalErrors / Convert.ToDouble(test_Data.Count)) * 100));
}
public WeightBias CalculateWB(WeightBias wb)
{
Dictionary <int, double> w = wb.Weight;
double b = wb.Bias;
int errors = wb.Updates;
foreach (var item in Training_Data)
{
int y = item.Sign; // true label
Dictionary <int, double> x = item.Vector;
double xw = 0;
foreach (var xi in x)
{
if (w.ContainsKey(xi.Key))
{
xw = xw + (w[xi.Key] * xi.Value);
}
}
xw += b;
if (Logistic_Regression) //Logistic Regression
{
foreach (var xi in x) //foreach (KeyValuePair<int, double> wi in w) //update this 16 here.
{
if (w.ContainsKey(xi.Key)) //if contains key
{
w[xi.Key] = ((1 - (2 * Learning_Rate / Tradeoff)) * w[xi.Key]) + ((Learning_Rate * y * xi.Value) / (Math.Exp(y * xw) + 1));
}
else //if doesn't contain key, it would x[wi.Key] would result to 0, so:
{
w[xi.Key] = ((1 - (2 * Learning_Rate / Tradeoff)) * RandomNumber()) + ((Learning_Rate * y * xi.Value) / (Math.Exp(y * xw) + 1));
}
}
b = ((1 - (2 * Learning_Rate / Tradeoff)) * b) + ((Learning_Rate * y) / (Math.Exp(y * b) + 1));
errors++;
}
else //Support Vector Machine (SVM)
{
if (y * xw <= 1)
{
foreach (var xi in x) //foreach (KeyValuePair<int, double> wi in w) //update this 16 here.
{
if (w.ContainsKey(xi.Key)) //if contains key
{
w[xi.Key] = ((1 - Learning_Rate) * w[xi.Key]) + (Learning_Rate * C * y * xi.Value);
}
else //if doesn't contain key, it would x[wi.Key] would result to 0, so:
{
w.Add(xi.Key, ((1 - Learning_Rate) * RandomNumber()) + (Learning_Rate * C * y * xi.Value));
}
}
b = ((1 - Learning_Rate) * b) + (Learning_Rate * C * y);
errors++;
}
else
{
foreach (var xi in x)
{
if (w.ContainsKey(xi.Key))
{
w[xi.Key] = ((1 - Learning_Rate) * w[xi.Key]);
}
else
{
w.Add(xi.Key, ((1 - Learning_Rate) * RandomNumber()));
}
}
b = ((1 - Learning_Rate) * b);
}
}
}
return(new WeightBias(w, b, errors));
}
public Data(int epochs, double learning_rate, double margin, double c, bool logistic_regression, double tradeoff, Random r,
List <Entry> train, List <Entry> test)
{
double temp_accuracy1;
double temp_accuracy2;
double temp_accuracy3;
double temp_accuracy4;
double temp_accuracy5;
Learning_Rate = learning_rate;
C = c;
Tradeoff = tradeoff;
Training_Data = new List <Entry>();
Test_Data = new List <Entry>();
Cross_Validate_Data = train.Concat(test).ToList();
Cross_1 = new List <Entry>();
Cross_2 = new List <Entry>();
Cross_3 = new List <Entry>();
Cross_4 = new List <Entry>();
Cross_5 = new List <Entry>();
SetValidateData(r);
#region First Fold
Training_Data = new List <Entry>();
Test_Data = new List <Entry>();
Training_Data = Cross_1.Concat(Cross_2.Concat(Cross_3.Concat(Cross_4))).ToList();
Test_Data = Cross_5;
perceptron = new Perceptron(Training_Data, Test_Data, learning_rate, margin, C, logistic_regression, Tradeoff, r);
Dictionary <int, double> w = new Dictionary <int, double>();
double b = (r.NextDouble() * (0.01 + 0.01) - 0.01);
WeightBias wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy1 = perceptron.GetAccuracy(Test_Data, wb);
#endregion
#region Second Fold
Training_Data = new List <Entry>();
Test_Data = new List <Entry>();
Training_Data = Cross_1.Concat(Cross_2.Concat(Cross_3.Concat(Cross_5))).ToList();
Test_Data = Cross_4;
perceptron = new Perceptron(Training_Data, Test_Data, learning_rate, margin, C, logistic_regression, Tradeoff, r);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy2 = perceptron.GetAccuracy(Test_Data, wb);
#endregion
#region Third Fold
Training_Data = new List <Entry>();
Test_Data = new List <Entry>();
Training_Data = Cross_1.Concat(Cross_2.Concat(Cross_4.Concat(Cross_5))).ToList();
Test_Data = Cross_3;
perceptron = new Perceptron(Training_Data, Test_Data, learning_rate, margin, C, logistic_regression, Tradeoff, r);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy3 = perceptron.GetAccuracy(Test_Data, wb);
#endregion
#region Fourth Fold
Training_Data = new List <Entry>();
Test_Data = new List <Entry>();
Training_Data = Cross_1.Concat(Cross_3.Concat(Cross_4.Concat(Cross_5))).ToList();
Test_Data = Cross_2;
perceptron = new Perceptron(Training_Data, Test_Data, learning_rate, margin, C, logistic_regression, Tradeoff, r);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy4 = perceptron.GetAccuracy(Test_Data, wb);
#endregion
#region Fifth Fold
Training_Data = new List <Entry>();
Test_Data = new List <Entry>();
Training_Data = Cross_2.Concat(Cross_3.Concat(Cross_4.Concat(Cross_5))).ToList();
Test_Data = Cross_1;
perceptron = new Perceptron(Training_Data, Test_Data, learning_rate, margin, C, logistic_regression, Tradeoff, r);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy5 = perceptron.GetAccuracy(Test_Data, wb);
#endregion
Test_Accuracy = (temp_accuracy1 + temp_accuracy2 + temp_accuracy3 + temp_accuracy4 + temp_accuracy5) / 5;
}
public WeightBias CalculateWB(WeightBias wb)
{
double[] w = wb.Weight;
double b = wb.Bias;
int updates = wb.Updates;
foreach (var item in Training_Data)
{
if (DymanicLearningRate)
{
Learning_Rate = Initial_Learning_Rate / T_Count;
}
int y = item.Sign;
int yguess;
double[] x = item.Vector;
double xw = 0;
for (int i = 0; i < ForestSize; i++)
{
xw = xw + (x[i] * w[i]);
}
xw += b;
if (Logistic_Regression) //Logistic Regression
{
for (int i = 0; i < ForestSize; i++) //(var xi in x) //foreach (KeyValuePair<int, double> wi in w) //update this ForestSize here.
{
if (x[i] != 0)
{
w[i] = ((1 - (2 * Learning_Rate / Tradeoff)) * w[i]) + ((Learning_Rate * y * x[i]) / (Math.Exp(y * xw) + 1));
}
}
b = ((1 - (2 * Learning_Rate / Tradeoff)) * b) + ((Learning_Rate * y) / (Math.Exp(y * b) + 1));
updates++;
}
else if (SVM)
{
if (y * xw <= 1)
{
for (int i = 0; i < ForestSize; i++) //foreach (KeyValuePair<int, double> wi in w) //update this ForestSize here.
{
if (x[i] != 0)
{
w[i] = ((1 - Learning_Rate) * w[i]) + (Learning_Rate * C * y * x[i]);
}
}
b = ((1 - Learning_Rate) * b) + (Learning_Rate * C * y);
updates++;
}
else
{
for (int i = 0; i < ForestSize; i++) //foreach (var xi in x)
{
if (x[i] != 0)
{
w[i] = ((1 - Learning_Rate) * w[i]);
}
}
b = ((1 - Learning_Rate) * b);
}
}
else //Perceptron
{
if (xw >= Margin)
{
yguess = +1;
}
else
{
yguess = -1;
}
if (y != yguess)
{
if (Aggressive)
{
double rhs = y * xw;
double top = Margin - rhs;
double xx = 0;
for (int i = 0; i < ForestSize; i++)
{
xx = xx + (x[i] * x[i]);
}
xx++;
Learning_Rate = top / xx;
for (int i = 0; i < ForestSize; i++)
{
w[i] = w[i] + (Learning_Rate * y * x[i]);
}
b = b + (Learning_Rate * y);
}
else
{
for (int i = 0; i < ForestSize; i++)
{
w[i] = w[i] + (Learning_Rate * y * x[i]);
}
b = b + (Learning_Rate * y);
}
updates++;
}
if (DymanicLearningRate)
{
T_Count++;
}
if (WeightBias_Average != null)
{
for (int i = 0; i < ForestSize; i++)
{
WeightBias_Average.Weight[i] += w[i];
}
WeightBias_Average.Bias += b;
}
}
}
return(new WeightBias(w, b, updates));
}
public AccuracyWB(double accuracy, WeightBias weight_Bias)
{
Weight_Bias = weight_Bias;
Accuracy = accuracy;
}
public Data(List <Entry> train, List <Entry> test, int epochs, double learning_rate, Random r, bool DymanicLearningRate, double margin, bool Average, bool Aggressive,
double c, bool svm, double tradeoff, bool logistic_regression, int forestSize)
{
C = c;
SVM = svm;
Tradeoff = tradeoff;
Logistic_Regression = logistic_regression;
ForestSize = forestSize;
double[] w_average = new double[ForestSize];
double b_average;
WeightBias wb_average = null;
if (Average)
{
for (int i = 0; i < ForestSize; i++)
{
double randomNumber = (r.NextDouble() * (0.01 + 0.01) - 0.01);
w_average[i] = randomNumber;
}
b_average = (r.NextDouble() * (0.01 + 0.01) - 0.01);
wb_average = new WeightBias(w_average, b_average, 0);
}
data_1 = train;
data_2 = test;
AccuracyWeightB = new Dictionary <int, AccuracyWB>();
perceptron = new Perceptron(data_1, data_2, learning_rate, DymanicLearningRate, margin, wb_average, Aggressive, C, SVM, Tradeoff, Logistic_Regression, ForestSize);
double[] w = new double[ForestSize];
double b = (r.NextDouble() * (0.01 + 0.01) - 0.01);
for (int i = 0; i < ForestSize; i++)
{
double randomNumber = (r.NextDouble() * (0.01 + 0.01) - 0.01);
w[i] = randomNumber;
}
WeightBias wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
if (Average)
{
perceptron.WeightBias_Average.Updates = wb.Updates;
AccuracyWeightB.Add(i + 1, new AccuracyWB(perceptron.GetAccuracy(data_2, perceptron.WeightBias_Average), perceptron.WeightBias_Average));
}
else
{
AccuracyWeightB.Add(i + 1, new AccuracyWB(perceptron.GetAccuracy(data_2, wb), wb));
}
perceptron.ShuffleTraining_Data(r);
}
//foreach (var item in AccuracyWeightB)
//{
// Console.WriteLine(item.Value.Accuracy);
//}
AccuracyWB bestAccuracy = AccuracyWeightB.OrderByDescending(x => x.Value.Accuracy).ThenByDescending(y => y.Key).Select(z => z.Value).First();
Test_Accuracy = bestAccuracy.Accuracy;
BestWeightBias = bestAccuracy.Weight_Bias;
Learning_Rate = learning_rate;
//Console.WriteLine("\n" + Accuracy);
}
public Data(int epochs, double learning_rate, Random r, bool DymanicLearningRate, double margin, bool Average, bool Aggressive,
double c, bool svm, double tradeoff, bool logistic_regression, List <Entry> train, List <Entry> test, int forestSize)
{
double temp_accuracy1;
double temp_accuracy2;
double temp_accuracy3;
double temp_accuracy4;
double temp_accuracy5;
data_1 = new List <Entry>();
data_2 = new List <Entry>();
Cross_Validate_Data = train.Concat(test).ToList();
Cross_1 = new List <Entry>();
Cross_2 = new List <Entry>();
Cross_3 = new List <Entry>();
Cross_4 = new List <Entry>();
Cross_5 = new List <Entry>();
SetValidateData(null, null, r);
C = c;
SVM = svm;
Tradeoff = tradeoff;
Logistic_Regression = logistic_regression;
Learning_Rate = learning_rate;
Margin = margin;
ForestSize = forestSize;
double[] w_average = new double[ForestSize];
double b_average;
WeightBias wb_average = null;
if (Average)
{
for (int i = 0; i < ForestSize; i++)
{
double randomNumber = (r.NextDouble() * (0.01 + 0.01) - 0.01);
w_average[i] = randomNumber;
}
b_average = (r.NextDouble() * (0.01 + 0.01) - 0.01);
wb_average = new WeightBias(w_average, b_average, 0);
}
#region First Fold
data_1 = Cross_1.Concat(Cross_2.Concat(Cross_3.Concat(Cross_4))).ToList();
data_2 = Cross_5;
perceptron = new Perceptron(data_1, data_2, learning_rate, DymanicLearningRate, margin, wb_average, Aggressive, C, SVM, Tradeoff, Logistic_Regression, ForestSize);
double[] w = new double[ForestSize];
double b = (r.NextDouble() * (0.01 + 0.01) - 0.01);
for (int i = 0; i < ForestSize; i++)
{
double randomNumber = (r.NextDouble() * (0.01 + 0.01) - 0.01);
w[i] = randomNumber;
}
WeightBias wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy1 = perceptron.GetAccuracy(data_2, wb);
if (Average)
{
temp_accuracy1 = perceptron.GetAccuracy(data_2, perceptron.WeightBias_Average);
}
#endregion
#region Second Fold
data_1 = new List <Entry>();
data_2 = new List <Entry>();
data_1 = Cross_1.Concat(Cross_2.Concat(Cross_3.Concat(Cross_5))).ToList();
data_2 = Cross_4;
perceptron = new Perceptron(data_1, data_2, learning_rate, DymanicLearningRate, margin, wb_average, Aggressive, C, SVM, Tradeoff, Logistic_Regression, ForestSize);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy2 = perceptron.GetAccuracy(data_2, wb);
if (Average)
{
temp_accuracy2 = perceptron.GetAccuracy(data_2, perceptron.WeightBias_Average);
}
#endregion
#region Third Fold
data_1 = new List <Entry>();
data_2 = new List <Entry>();
data_1 = Cross_1.Concat(Cross_2.Concat(Cross_4.Concat(Cross_5))).ToList();
data_2 = Cross_3;
perceptron = new Perceptron(data_1, data_2, learning_rate, DymanicLearningRate, margin, wb_average, Aggressive, C, SVM, Tradeoff, Logistic_Regression, ForestSize);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy3 = perceptron.GetAccuracy(data_2, wb);
if (Average)
{
temp_accuracy3 = perceptron.GetAccuracy(data_2, perceptron.WeightBias_Average);
}
#endregion
#region Fourth Fold
data_1 = new List <Entry>();
data_2 = new List <Entry>();
data_1 = Cross_1.Concat(Cross_3.Concat(Cross_4.Concat(Cross_5))).ToList();
data_2 = Cross_2;
perceptron = new Perceptron(data_1, data_2, learning_rate, DymanicLearningRate, margin, wb_average, Aggressive, C, SVM, Tradeoff, Logistic_Regression, ForestSize);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy4 = perceptron.GetAccuracy(data_2, wb);
if (Average)
{
temp_accuracy4 = perceptron.GetAccuracy(data_2, perceptron.WeightBias_Average);
}
#endregion
#region Fifth Fold
data_1 = new List <Entry>();
data_2 = new List <Entry>();
data_1 = Cross_2.Concat(Cross_3.Concat(Cross_4.Concat(Cross_5))).ToList();
data_2 = Cross_1;
perceptron = new Perceptron(data_1, data_2, learning_rate, DymanicLearningRate, margin, wb_average, Aggressive, C, SVM, Tradeoff, Logistic_Regression, ForestSize);
wb = new WeightBias(w, b, 0);
for (int i = 0; i < epochs; i++)
{
wb = perceptron.CalculateWB(wb);
perceptron.ShuffleTraining_Data(r);
}
temp_accuracy5 = perceptron.GetAccuracy(data_2, wb);
if (Average)
{
temp_accuracy5 = perceptron.GetAccuracy(data_2, perceptron.WeightBias_Average);
}
#endregion
Test_Accuracy = (temp_accuracy1 + temp_accuracy2 + temp_accuracy3 + temp_accuracy4 + temp_accuracy5) / 5;
}