public void Assign(NEAT source)
{
foreach (var n in nodes)
{
Destroy(n);
}
nodes.Clear();
foreach (var c in connections)
{
Destroy(c.gameObject);
}
connections.Clear();
source.InputLayer.ForEach(neuron => neuron.Done = false);
source.OutputLayer.ForEach(neuron => neuron.Done = false);
source.HiddenLayers.Values.ToList().ForEach(neuron => neuron.Done = false);
horizontal = 1;
i = 0;
var position = initialPosittion;
foreach (var neuron in source.InputLayer)
{
ComputeNeuron(neuron, position);
position.y += offset;
}
position = initialPosittion;
position.x += offset * (source.HiddenLayers.Count + 1) * 2;
position.y += offset * (source.InputLayer.Count - source.OutputLayer.Count) / 2;
foreach (var neuron in source.OutputLayer)
{
ComputeNeuron(neuron, position);
position.y += offset;
}
}
public float Distance(NEAT partner)
{
if (partner == null)
{
Debug.Log("NULL partner");
Debug.Break();
}
//var synapseIenumerator = AllSynapses.GetEnumerator();
//var partnerSynapseIenumerator = partner.AllSynapses.GetEnumerator();
//synapseIenumerator.MoveNext();
float result = 0;
if (AllSynapses.Any() || partner.AllSynapses.Any())
{
int disjointGenes = 0;
int excessGenes = 0;
float weightDifference = 0;
float N = Math.Max(AllSynapses.Count, partner.AllSynapses.Count);
disjointGenes = 2 * AllSynapses.Keys.Union(partner.AllSynapses.Keys).Count() - AllSynapses.Count - partner.AllSynapses.Count;
foreach (var synapseInnovationNo in AllSynapses.Keys.Intersect(partner.AllSynapses.Keys))
{
weightDifference += Math.Abs((float)(AllSynapses[synapseInnovationNo].Weight - partner.AllSynapses[synapseInnovationNo].Weight));
//if (Math.Abs((float)(AllSynapses[synapseInnovationNo].Weight - partner.AllSynapses[synapseInnovationNo].Weight)) > Constants.Con.synapse_difference_threshold)
// disjointGenes++;
}
result += (Constants.Con.c1 * excessGenes + Constants.Con.c2 * disjointGenes + Constants.Con.c3 * weightDifference) / N;
}
//if(HiddenLayers.Any() || partner.HiddenLayers.Any()) OutputLayer.Count > 0 always
{
int disjointGenes = 0;
int excessGenes = 0;
float weightDifference = 0;
float N = Math.Max(HiddenLayers.Count, partner.HiddenLayers.Count) + OutputLayer.Count;
foreach (var neuronPair in OutputLayer.Zip(partner.OutputLayer))
{
weightDifference += Mathf.Abs((float)(neuronPair.Key.Bias - neuronPair.Value.Bias));
//if (Mathf.Abs((float)(neuronPair.Key.Bias - neuronPair.Value.Bias)) > Constants.Con.bias_difference_threshold)
// disjointGenes++;
}
disjointGenes = 2 * HiddenLayers.Keys.Union(partner.HiddenLayers.Keys).Count() - HiddenLayers.Count - partner.HiddenLayers.Count;
foreach (var neuronInnovationNo in HiddenLayers.Keys.Intersect(partner.HiddenLayers.Keys))
{
weightDifference += Math.Abs((float)(HiddenLayers[neuronInnovationNo].Bias - partner.HiddenLayers[neuronInnovationNo].Bias));
//if (Math.Abs((float)(HiddenLayers[neuronInnovationNo].Bias - partner.HiddenLayers[neuronInnovationNo].Bias)) > Constants.Con.bias_difference_threshold)
// disjointGenes++;
}
result += (Constants.Con.c1 * excessGenes + Constants.Con.c2 * disjointGenes + Constants.Con.c3 * weightDifference) / N;
}
return(result);
//if (result < Constants.Con.delta_t)
// return true;
//return false;
}
public NEAT Crossover(NEAT partner)
{
//assume: partner has higher score
NEAT offspring = partner.Copy();
foreach (var synapse in AllSynapses)
{
if (offspring.AllSynapses.ContainsKey(synapse.Key) && RandomGenerator.Next(2) == 0)
{
offspring.AllSynapses[synapse.Key].Weight = synapse.Value.Weight;
}
}
return(offspring);
}
public NEAT Copy()
{
NEAT c = new NEAT(InputLayer.Count, OutputLayer.Count, true);
foreach (var synapse in AllSynapses)
{
c.AddSynapse(synapse.Value);
}
foreach (var neuronpair in OutputLayer.Zip(c.OutputLayer))
{
neuronpair.Value.Bias = neuronpair.Key.Bias;
}
return(c);
}
