/// <summary>
/// Returns the quadratic cost of the test case for all inputs using:
/// Cost = ((Expected - Actual)^2) ==> Take the average of all costs.
/// </summary>
/// <param name="testCase">The training iteration to calculate the cost for.</param>
/// <returns>Returns the cost of the training iteration.</returns>
private double CalculateCost(INetworkTrainingIteration testCase)
{
var neuronOutputs = GetNeuronOutputs();
testCase.Outputs.ForEach(t =>
{
// Update the actual activation level of each test case.
t.ActivationLevel = neuronOutputs.First(n => n.NeuronId == t.NeuronId).ActivationLevel;
});
return(testCase.Outputs.Sum(t => Math.Pow(t.ActivationLevel - t.ExpectedActivationLevel, 2)));
}
/// <summary>
/// Creates and returns a neural network using configuration settings in the <paramref name="networkConfiguration"/> and the <paramref name="trainingEntry"/>.
/// </summary>
/// <param name="networkConfiguration"></param>
/// <param name="trainingDatasetEntries"></param>
/// <returns>Returns a neural network using configuration settings in the <paramref name="networkConfiguration"/> and the <paramref name="trainingEntry"/>.</returns>
private IDFFNeuralNetwork CreateNetwork(N networkConfiguration, INetworkTrainingIteration trainingEntry)
{
if (trainingEntry == null || trainingEntry.Inputs == null || trainingEntry.Outputs == null)
{
throw new ArgumentNullException("trainingEntry");
}
var numInputs = trainingEntry.Inputs.Count();
var numOutputs = trainingEntry.Outputs.Count();
var numHiddenLayers = networkConfiguration.NumHiddenLayers;
var numHiddenLayerNeurons = networkConfiguration.NumHiddenLayerNeurons;
// Setup and randomize network.
var network = new DFFNeuralNetwork(numInputs, numHiddenLayers, numHiddenLayerNeurons, numOutputs);
network.RandomizeNetwork();
return(network);
}
