private void StartTraining()
{
var(inputData, expectedOutputData) = NeuralNetworkHelper.ReduceDataset(
humanCar.StateInputMeasurements.ToList(),
humanCar.StateOutputMeasurements.ToList());
for (int i = 0; i < inputData.Count(); i++)
{
if (inputData[i].All(d => d == 0))
{
inputData.RemoveAt(i);
expectedOutputData.RemoveAt(i);
i--;
}
}
notificationService.ShowToast(
ToastType.Info,
"Training Started...");
// Right now the training happens so fast, not sure if we need to be reporting progress
network = NeuralNetworkHelper.GetTrainedNetwork(inputData, expectedOutputData, (c, t) => { });
notificationService.ShowToast(
ToastType.Info,
"Training Complete.");
}
public void Initialize(Track track, MLPNeuralNetwork controller)
{
var carAi = new CarAI(random, controller);
// Load the simulation
simulation = appService.Kernel.Get <RacingSimulationLogic>();
simulation.SetTrack(track);
// Add the 1:1 trained AI to the list of cars
var cars = new List <ICarController>()
{
carAi
};
// Spawn 100 mutated versions of our AI
for (int i = 0; i < 100; i++)
{
var car = carAi.Clone();
car.Mutate();
cars.Add((CarAI)car);
}
// Initialize the simulation
simulation.SetCars(cars);
simulation.ResetCars();
// If the visualization is turned on, create it, set the track and add the cars.
simulationVisualization = appService.Kernel.Get <RacingSimulationVisualization>();
simulationVisualization.SetTrack(track);
simulationVisualization.InitializeCars(simulation.GetCars());
}
public CarAI(Random random, MLPNeuralNetwork network)
{
this.networkStructure = network.GetStructure();
this.Network = network;
this.Random = random;
this.Configuration = new CarConfiguration();
}
private void TrainNetwork()
{
// Define the layour for our NN.
// -- Input Nodes 784 - 1 per pixel (28x28).
// -- 2 Hidden Layer. 16 nodes each - arbitrary.
// -- 1 Output Layer. 10 Nodes (0-10).
var inputNodes = 784;
var outputNodes = 10;
var innerLayers = 2;
var hiddenLayerDepth = 16;
// Inner Layer + Input + Output = 3
var layers = new int[innerLayers + 2];
layers[0] = inputNodes;
for (int i = 1; i < innerLayers + 1; i++)
{
layers[i] = hiddenLayerDepth;
}
layers[innerLayers + 1] = outputNodes;
// Create the NN.
net = new MLPNeuralNetwork(layers);
// Grab the training data from the images loaded.
var trainData = imageExtraction.Images.Select(i => i.GetImageAsInput()).ToList();
// Grab the reinforcement data from the labels provided.
var reinforceValues = imageExtraction.Images.Select(i => i.integerValues).ToList();
// Train the network with our known good data.
net.Train(trainData, reinforceValues, ReportProgress);
// We now want to display our guess of the number.
guessNumber = true;
}
