public Neuron(Neuron oldNeuron)
{
this.activationFunction = oldNeuron.activationFunction;
this.activationFunctionDerivative = oldNeuron.activationFunctionDerivative;
this.weights = new double[oldNeuron.weights.Length];
oldNeuron.weights.CopyTo(this.weights, 0);
}
public Layer(int numberOfInputs, int numberOfNeurons, ActivationFunction f, ActivationFunctionDerivative g)
{
this.neurons = new Neuron[numberOfNeurons];
for (int i = 0; i < numberOfNeurons; i++)
{
this.neurons[i] = new Neuron(numberOfInputs, f, g);
}
}
public Neuron(int weights, ActivationFunction activationFunction, ActivationFunctionDerivative activationFunctionDerivative)
{
Random rnd = new Random();
this.weights = new double[weights];
for (int i = 0; i < weights; i++)
{
this.weights[i] = rnd.NextDouble() * 2 - 1;
}
this.activationFunction = activationFunction;
this.activationFunctionDerivative = activationFunctionDerivative;
this.lastValue = 0;
}
public double ActivateDerivative(double activity, double output)
{
double fe = FlatspotEliminationEnable.Value ? FlatspotElimination.Value : 0;
EActivationType at = ActivationType.Value;
switch (at)
{
case EActivationType.Symmetric:
return(fe + 1 - output * output);
case EActivationType.Asymmetric:
return(fe + output * (1 - output));
case EActivationType.Threshold:
return(activity == ActivationThreshold.Value ? double.PositiveInfinity : fe); //TODO: hardly useful ...
case EActivationType.Userdefined:
return(fe + ActivationFunctionDerivative.Value(activity, output));
default: //case EActivationType.Linear:
return(1);
}
}
public LayerParams(ActivationFunction f, ActivationFunctionDerivative g, int n)
{
this.F = f;
this.G = g;
this.NumNeurons = n;
}
static void Main(string[] args)
{
//Activation functions
ActivationFunction sigmoidFunc = (x) => 1 / (1 + Math.Exp(-x));
ActivationFunctionDerivative sigmoidFuncDerivative = (x) => x * (1 - x);
ActivationFunction reluFunc = (x) => Math.Max(0, x);
ActivationFunctionDerivative reluFuncDerivative = (x) => { if (x < 0)
{
return(0);
}
else
{
return(1);
} };
double eluParam = 0.001;
ActivationFunction eluFunc = (x) => { if (x > 0)
{
return(x);
}
else
{
return(eluParam * (Math.Exp(x) - 1));
} };
ActivationFunctionDerivative eluFuncDerivative = (x) => { if (x < 0)
{
return(x * eluParam * Math.Exp(x - 1));
}
else
{
return(1);
} };
bool isEA = true;
DataStruct[] trainingSet = new DataStruct[4];
trainingSet[0] = new DataStruct(new double[] { 1.0, 1.0 }, new double[] { 0.0, 1.0 });
trainingSet[1] = new DataStruct(new double[] { -1.0, 1.0 }, new double[] { 1.0, 0.0 });
trainingSet[2] = new DataStruct(new double[] { 1.0, -1.0 }, new double[] { 1.0, 0.0 });
trainingSet[3] = new DataStruct(new double[] { -1.0, -1.0 }, new double[] { 0.0, 1.0 });
if (!isEA)
{
var network = new Network(2);
network.AddLayer(sigmoidFunc, sigmoidFuncDerivative, 10);
network.AddLayer(eluFunc, eluFuncDerivative, 100);
network.AddLayer(sigmoidFunc, sigmoidFuncDerivative, 10);
network.AddLayer(eluFunc, eluFuncDerivative, 100);
network.AddLayer(sigmoidFunc, sigmoidFuncDerivative, 10);
network.AddLayer(eluFunc, eluFuncDerivative, 100);
network.AddLayer(sigmoidFunc, sigmoidFuncDerivative, 10);
network.AddLayer(sigmoidFunc, sigmoidFuncDerivative, 2);
network.BuildNetwork();
network.Train(trainingSet, 0.9, 100000, 1000);
Console.WriteLine("1 and 1");
var res = network.Inference(trainingSet[0].Input);
for (int i = 0; i < res.Length; i++)
{
Console.Write($"{res[i]} ");
}
Console.WriteLine("");
Console.WriteLine("1 and 0");
res = network.Inference(trainingSet[2].Input);
for (int i = 0; i < res.Length; i++)
{
Console.Write($"{res[i]} ");
}
Console.WriteLine("");
Console.WriteLine("0 and 1");
res = network.Inference(trainingSet[1].Input);
for (int i = 0; i < res.Length; i++)
{
Console.Write($"{res[i]} ");
}
Console.WriteLine("");
Console.WriteLine("0 and 0");
res = network.Inference(trainingSet[3].Input);
for (int i = 0; i < res.Length; i++)
{
Console.Write($"{res[i]} ");
}
}
else
{
Game2048 game = new Game2048();
game.PrintBoard();
Console.WriteLine($"{game.Score}");
game.Play(0);
game.PrintBoard();
Console.WriteLine($"{game.Score}");
game.Play(1);
game.PrintBoard();
Console.WriteLine($"{game.Score}");
game.Play(2);
game.PrintBoard();
Console.WriteLine($"{game.Score}");
game.Play(3);
game.PrintBoard();
Console.WriteLine($"{game.Score}");
//LayerParams[] layers = new LayerParams[3];
//layers[0] = new LayerParams(sigmoidFunc, sigmoidFuncDerivative, 10);
//layers[1] = new LayerParams(sigmoidFunc, sigmoidFuncDerivative, 10);
//layers[2] = new LayerParams(sigmoidFunc, sigmoidFuncDerivative, 2);
//EA ea = new EA(100, 2, layers);
//ea.EvolutionaryLoop(trainingSet, 10000, 90, 0.5, 0.7, 10, 0.3);
//var winner = ea.Population.OrderBy(x => x.Fitness).ThenBy(x => x.FromGeneration).ElementAt(0);
//Console.WriteLine($"The best individual was created in generation: {winner.FromGeneration} and went through {winner.NumMutations} mutations");
//Console.WriteLine("1 and 1");
//var res = winner.Individual.Inference(trainingSet[0].Input);
//for (int i = 0; i < res.Length; i++)
//{
// Console.Write($"{res[i]} ");
//}
//Console.WriteLine("");
//Console.WriteLine("1 and 0");
//res = winner.Individual.Inference(trainingSet[2].Input);
//for (int i = 0; i < res.Length; i++)
//{
// Console.Write($"{res[i]} ");
//}
//Console.WriteLine("");
//Console.WriteLine("0 and 1");
//res = winner.Individual.Inference(trainingSet[1].Input);
//for (int i = 0; i < res.Length; i++)
//{
// Console.Write($"{res[i]} ");
//}
//Console.WriteLine("");
//Console.WriteLine("0 and 0");
//res = winner.Individual.Inference(trainingSet[3].Input);
//for (int i = 0; i < res.Length; i++)
//{
// Console.Write($"{res[i]} ");
//}
}
}
public void AddLayer(ActivationFunction f, ActivationFunctionDerivative g, int numberOfNeurons)
{
this.LayerList.Add(new LayerParams(f, g, numberOfNeurons));
}
