//This is the default test for the neural network to confirm it is performing as expected
public string getTestDataString(NN n)
{
StringBuilder outStringBuilder = new StringBuilder();
outStringBuilder.AppendLine("Test Neural Network");
outStringBuilder.AppendLine("-------------------");
//Print the input data
outStringBuilder.AppendLine("Input Data");
outStringBuilder.AppendFormat("Number of Inputs: {0}\n", n.NumberOfInputs);
outStringBuilder.AppendFormat("Number of Outputs: {0}\n", n.NumberOfOutputs);
outStringBuilder.AppendFormat("Number of Hidden Layers: {0}\n", n.NumberOfHiddenLayers);
outStringBuilder.AppendLine("Number of Nodes Per Hidden Layer: ");
Array.ForEach(n.NumberOfNodesPerHiddenLayer, x => outStringBuilder.AppendFormat("{0} ", x));
outStringBuilder.AppendLine();
//Print the calculated data
outStringBuilder.AppendLine("-------------------");
outStringBuilder.AppendLine("General Calculated Metadata");
outStringBuilder.AppendFormat("Total Number of Layers: {0}\n", n.NumberOfLayers);
outStringBuilder.AppendFormat("Total Number of Nodes: {0}\n", n.NumberOfNodes);
outStringBuilder.AppendFormat("Total Number of Hidden Nodes: {0}\n", n.NumberOfHiddenNodes);
outStringBuilder.AppendLine();
//Print the layer data
outStringBuilder.AppendLine("-------------------");
outStringBuilder.AppendLine("Specific Calculated Metadata Per Layer");
outStringBuilder.AppendLine("Number of Weights Per Layer: ");
Array.ForEach(n.WeightsPerLayer, x => outStringBuilder.AppendFormat("{0} ", x));
outStringBuilder.AppendLine();
outStringBuilder.AppendLine("Number of Inputs Per Layer: ");
Array.ForEach(n.InputsPerLayer, x => outStringBuilder.AppendFormat("{0} ", x));
outStringBuilder.AppendLine();
outStringBuilder.AppendLine("Number of Nodes Per Layer: ");
Array.ForEach(n.NodesPerLayer, x => outStringBuilder.AppendFormat("{0} ", x));
outStringBuilder.AppendLine();
outStringBuilder.AppendLine("Weight Start Index Per Layer: ");
Array.ForEach(n.WeightStartIndexPerLayer, x => outStringBuilder.AppendFormat("{0} ", x));
outStringBuilder.AppendLine();
outStringBuilder.AppendLine("Weight Start Index Per Node Per Layer: ");
for (int i = 0; i < n.WeightStartIndexPerLayerPerNode.Length; i++)
{
Array.ForEach(n.WeightStartIndexPerLayerPerNode[i], x => outStringBuilder.AppendFormat("{0} ", x));
outStringBuilder.AppendLine();
}
outStringBuilder.AppendLine();
outStringBuilder.AppendLine();
//Print the detailed data
outStringBuilder.AppendLine("-------------------");
outStringBuilder.AppendLine("Detailed Data");
outStringBuilder.AppendFormat("Number of Weights: {0}\n", n.Weights.Length);
outStringBuilder.AppendLine("Weights: ");
Array.ForEach(n.Weights, x => outStringBuilder.AppendFormat("{0}\n", x));
outStringBuilder.AppendLine();
//Return the results string
return outStringBuilder.ToString();
}
public List<NN> BreedPickEach(Tuple<int, int>[] couples, KeyValuePair<int, double>[] sortedFitnessList)
{
//Create a list of new neural networks for the next generation
List<NN> newPopulation = new List<NN>();
//Place the elites in the population
for (int i = 0; i < numberOfElites; i++)
for (int j = 0; j < numOfCopiesOfElites; j++)
newPopulation.Add(population[sortedFitnessList[i].Key]);
//Filling the remainder of the population
int k = 0;
while (newPopulation.Count < population.Count)
{
//If we randomly are above the crossover rate
if (Util.randNumGen.NextDouble() > crossoverRate)
{
//Add the first parent
newPopulation.Add(population[couples[k].Item1]);
//If we have room for another add the second parent
if (!(newPopulation.Count <= population.Count))
newPopulation.Add(population[couples[k].Item2]);
}
//If we are below the crossover rate
else
{
//Create a new neural network with the same structure as the rest of the population
NN child1 = new NN(networkStructure[0], networkStructure[networkStructure.Length - 1], hiddenLayerStructure.Length, hiddenLayerStructure);
//Create a new neural network with the same structure as the rest of the population
NN child2 = new NN(networkStructure[0], networkStructure[networkStructure.Length - 1], hiddenLayerStructure.Length, hiddenLayerStructure);
//Add all genes from each parent up to the crossover point to the two different children
int i;
for (i = 0; i < population[k].Weights.Length; i++)
{
//Randomly stitch one of the parents genes onto each child
if (Util.randNumGen.NextDouble() > 0.5)
{
child1.Weights[i] = population[couples[k].Item1].Weights[i];
child2.Weights[i] = population[couples[k].Item2].Weights[i];
}
else
{
child1.Weights[i] = population[couples[k].Item2].Weights[i];
child2.Weights[i] = population[couples[k].Item1].Weights[i];
}
}
//Add the first child to the new population
newPopulation.Add(child1);
//If we have room for more, add the second child to the population
if (!(newPopulation.Count <= population.Count))
newPopulation.Add(child2);
}
//Confirm that the iteration variable does not exceed the array size from which it is pulling values
k = (k + 1) % couples.Length;
}
//Return the new population
return newPopulation;
}
//This function tests the serial functionality of the Neural Network object.
//This can be useful if a particularly great neural network is consturcted and should be saved/loaded.
private bool testSerialization(NN n)
{
//Save the neural network created when this object was built
n.SaveNNToFile("Test_NN.dat");
//Load a new neural network from that file
NN loadedNN = NN.LoadNNFromFile("Test_NN.dat");
//Using the custom equals function, test to see if the networks are equal then show the results.
if (loadedNN.Equals(n))
return true;
return false;
}
/// <summary>
/// Allows the game to perform any initialization it needs to before starting to run.
/// This is where it can query for any required services and load any non-graphic
/// related content. Calling base.Initialize will enumerate through any components
/// and initialize them as well.
/// </summary>
protected override void Initialize()
{
//Initialize a test network
testNN = new NN(4, 2, 1, new int[] { 6 });
//Build a test population
pop = new NNPopulation(simulationPopulationSize, neuralNetworkStructure, typeof(NNAntSimulation));
//Output the default object test results
///richTextBox_simpleOut.Text = getTestDataString();
//Fill the simThread with a default value (not running)
simThread = new Thread(new ThreadStart(runGeneticSimulation));
//No stop is request at the beginning of execution
requestStop = false;
///label_totalIterations.Text = "0 Total Iterations, Population Size: " + pop.StartPopulationSize;
///comboBox_BreedingType.SelectedIndex = 0;
prevKeyboardState = Keyboard.GetState();
prevMouseState = Mouse.GetState();
prevGamePadState = GamePad.GetState(PlayerIndex.One);
basicEffect = new BasicEffect(graphics.GraphicsDevice);
basicEffect.VertexColorEnabled = true;
basicEffect.Projection = Matrix.CreateOrthographicOffCenter
(0, graphics.GraphicsDevice.Viewport.Width, // left, right
graphics.GraphicsDevice.Viewport.Height, 0, // bottom, top
0, 1); // near, far plane
Util.genericTextFont = Content.Load<SpriteFont>("GenericFont");
base.Initialize();
}
