// Evaluate the algorithm using cross validation and return average score
public double CrossValidationEvaluation(double[][] X, double[] y, int k_folds, int n_epoch)
{
// Get the k fold validation split points
int[] dataset_split = new int[k_folds];
int fold_size;
(dataset_split, fold_size) = ML_tools.CrossValidationSplit(X, k_folds);
// Set up some more parameters
double[][] train_X;
double[] train_y;
double[][] test_X;
double[] test_y;
double score = 0;
double[] predictions = new double[fold_size];
// Run throught the algorithm k times
for (int k = 0; k < k_folds - 1; k++)
{
// Bit fiddly here to get the right data in the training set
if (!(k == 0))
{
List <double[]> list_X = new List <double[]>();
list_X.AddRange(ML_tools.Slice(X, dataset_split[0], dataset_split[k] - 1));
List <double> list_y = new List <double>();
list_y.AddRange(ML_tools.Slice(y, dataset_split[0], dataset_split[k] - 1));
if (!(k == k_folds - 1 || k == k_folds - 2))
{
list_X.AddRange(ML_tools.Slice(X, dataset_split[k + 1], dataset_split[k + 2] - 1));
list_y.AddRange(ML_tools.Slice(y, dataset_split[k + 1], dataset_split[k + 2] - 1));
}
train_X = list_X.ToArray();
train_y = list_y.ToArray();
}
else
{
train_X = ML_tools.Slice(X, dataset_split[k + 1], dataset_split[k + 2] - 1);
train_y = ML_tools.Slice(y, dataset_split[k + 1], dataset_split[k + 2] - 1);
}
test_X = ML_tools.Slice(X, dataset_split[k], dataset_split[k + 1] - 1);
test_y = ML_tools.Slice(y, dataset_split[k], dataset_split[k + 1] - 1);
// Train the network
Train(n_epoch, train_X, train_y);
// Make predictions with it
predictions = Predict(test_X);
// Find the ABSE and average it
score = ((score * k) + AbsoluteError(predictions, test_y)) / (k + 1);
}
return(score);
}
static void Main(string[] args)
{
// Saved data properties
string filename = "C:\\Users\\maxel\\OneDrive\\Machine learning\\Neural Networks\\Data_sets\\Boston_housing.csv";
int[] X_columns = new int[13] {
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 13
};
int y_columns = 11;
// Load the data
(double[][] X, double[] y) = ML_tools.ReadCSV(filename, X_columns, y_columns);
// Scale the data
ML_tools.MinMaxTransformation_X(X);
(double minimum, double range) = ML_tools.MinMaxTransformation_y(y);
// Neural network parameters
int hidden_layers = 2;
int[] hidden_nodes = new int[2] {
5, 3
}; // Must be the same length as the hidden layers
double l_rate_weights = 2;
double l_rate_bias = 2;
string activation_func = "sigmoid";
NeuralNetwork Neural_Net = new NeuralNetwork(hidden_layers, hidden_nodes, l_rate_weights, l_rate_bias, 13, 1, activation_func);
// Testing parameters
int n_epoch = 10000;
int k_folds = 5;
// Time how long it takes to execute
Stopwatch stopwatch = Stopwatch.StartNew();
// Use cross validation to evaluate the algorithm
Console.WriteLine("Cross validation absolute error: {0}", Neural_Net.CrossValidationEvaluation(X, y, k_folds, n_epoch));
stopwatch.Stop();
Console.WriteLine("Execution time: {0} seconds", (stopwatch.ElapsedMilliseconds / 1000));
Console.ReadLine();
}
