public static NeuralEnvironment RandomMerge(NeuralEnvironment first, NeuralEnvironment second)
{
var rand = new Random();
var newGene = first.Clone();
foreach (var neuron in second.GetNeurons())
{
var id = neuron.Id;
var newNeuron = newGene.NeuronsIds[id];
foreach (var(input, weight) in neuron.Inputs)
{
if (rand.NextDouble() > MergeCoeff)
{
newGene.Connect(newNeuron, newGene.NeuronsIds[input], weight);
}
if (rand.NextDouble() < MutationProbability)
{
newGene.Connect(newNeuron, newGene.NeuronsIds[input],
newNeuron.Inputs[input] + (rand.NextDouble() * 2 - 1));
}
//(rand.NextDouble() * 2 - 1));
}
}
return(newGene);
}
public static void Teach(NeuralEnvironment environment, Dictionary <List <double>, List <double> > data)
{
var count = 0;
var speed = 0.7;
foreach (var(example, answers) in data)
{
environment.Work(example);
var miss = new Dictionary <Neuron, (double, int)>();
var nextQue = new Dictionary <Neuron, (double, int)>();
var que = environment.GetOutputs()
.Zip(answers, (n, a) => (n, a))
.ToDictionary(t => t.n, t => ((t.a - t.n.Work(environment.NeuronsIds)), 1));
while (que.Count != 0)
{
foreach (var(neuron, (sum, num)) in que)
{
var res = neuron.Work(environment.NeuronsIds);
var q = sum / num;
var neuronQ = q * NetworkManager.GetDerivativeFunc(neuron.ActFunc)(res);
foreach (var(input, weight) in neuron.Inputs.ToList())
{
var inputNeuron = environment.NeuronsIds[input];
var delta = speed * neuronQ * inputNeuron.Work(environment.NeuronsIds);
if (nextQue.ContainsKey(inputNeuron))
{
nextQue[inputNeuron]
= (nextQue[inputNeuron].Item1 + neuronQ * weight, nextQue[inputNeuron].Item2 + 1);
}
else
{
nextQue.Add(inputNeuron, (neuronQ * weight, 1));
}
neuron.Connect(input, weight + delta);
}
}
que = nextQue;
nextQue = new Dictionary <Neuron, (double, int)>();
}
}
}
