public double[] Train(double[] inputs, double trainSpeed = 0.05)
{
if (inputs.Length != InputNeurons.Count)
{
throw new ArgumentException();
}
for (int i = 0; i < InputNeurons.Count; i++)
{
InputNeurons[i].Value = inputs[i];
}
Parallel.ForEach(OutputNeurons, x => x.Calculate());
var winner =
OutputNeurons.Select(
x =>
new
{
Neuron = x,
Value =
x.Dendrites.Sum(item => Math.Abs(item.Neuron.Value - item.Weight)) *
(Wins.ContainsKey(x) ? Wins[x] : 1)
})
.OrderBy(x => x.Value)
.First();
Wins[winner.Neuron] = Wins.ContainsKey(winner.Neuron) ? Wins[winner.Neuron] + 1 : 2;
foreach (var dendrite in winner.Neuron.Dendrites)
{
dendrite.Weight = dendrite.Weight + trainSpeed * (dendrite.Neuron.Value - dendrite.Weight);
}
return(OutputNeurons.Select(x => x.Value).ToArray());
}
public void AddOutputNeuronAndMesh(WorkingNeuron neuron)
{
OutputNeurons.Add(neuron);
foreach (WorkingNeuron wn in LastHiddenLayer)
{
neuron.AddNeuronConnection(new Connection(wn, 0));
}
}
public Genome(Genome target, List <Gene> otherGenes = null)
{
SpeciesId = target.SpeciesId;
Fitness = target.Fitness;
CreateNetwork(target.InputCount, target.OutputCount, otherGenes ?? target.Genes);
Assert.AreEqual(InputNeurons.Count(), target.InputCount);
Assert.AreEqual(OutputNeurons.Count(), target.OutputCount);
}
public IEnumerable <double> CalculateOutputs()
{
foreach (var cachedNeuron in AllNeurons.OfType <ACachedNeuron>())
{
cachedNeuron.ResetCache();
}
return(OutputNeurons.Select(o => o.Value));
}
/// <summary>
/// Saves the neural network as a .ann file.
/// </summary>
/// <param name="name">The .ann file name.</param>
/// <returns>The .ann file.</returns>
public AnnFile Save(string name)
{
var map = ComputeNodeIdMap();
return(new AnnFile
{
Name = name,
Activation = Activation,
InputNeurons = InputNeurons.Select(neuron => (uint)map[neuron.ID]).ToList(),
OutputNeurons = OutputNeurons.Select(neuron => (uint)map[neuron.ID]).ToList(),
AdjacencyMatrix = AdjacencyMatrix
});
}
public double[] Process(double[] inputs)
{
if (inputs.Length != InputNeurons.Count)
{
throw new ArgumentException();
}
for (int i = 0; i < InputNeurons.Count; i++)
{
InputNeurons[i].Value = inputs[i];
}
foreach (var outputNeuron in OutputNeurons)
{
outputNeuron.Calculate();
}
return(OutputNeurons.Select(x => x.Value).ToArray());
}
public Genome(int inputCount, int outputCount, float geneWeightRandomVal = 0)
{
if (inputCount < 1 || outputCount < 1)
{
throw new System.ArgumentException("Invalid initialization", "inputCount or outputCount");
}
SpeciesId = null;
Fitness = 0;
InitStartingNeurons(inputCount, outputCount);
InitStartingGenes(geneWeightRandomVal);
Assert.AreEqual(InputNeurons.Count(), inputCount);
Assert.AreEqual(OutputNeurons.Count(), outputCount);
Assert.AreEqual(Neurons.Count, inputCount + outputCount + 1);
Assert.AreEqual(Genes.Count, inputCount * outputCount);
}
/// <summary>
/// Create a network from the given genes without any prerequesties.
/// </summary>
/// <param name="targetGenes"></param>
private void CreateNetwork(int inputCount, int outputCount, IEnumerable <Gene> targetGenes)
{
Neurons = new List <Neuron>();
Genes = new List <Gene>(targetGenes.Count());
AddStartingNeurons(inputCount, outputCount);
foreach (var gene in targetGenes)
{
Genes.Add(ForceConnection(gene));
}
InputNeurons = Neurons.Where(x => x.Type == ENeuronType.Input);
OutputNeurons = Neurons.Where(x => x.Type == ENeuronType.Output);
Debug.Assert(Genes.Count == targetGenes.Count());
Debug.Assert(InputNeurons.Count() == inputCount);
Debug.Assert(OutputNeurons.Count() == outputCount);
}
/// <summary>
/// Initialize the network with the starting genes.
/// All inputs, but the bias, will be connected to output neurons.
/// </summary>
private void InitStartingGenes(float geneWeightRandomVal)
{
Genes = new List <Gene>(InputCount * OutputCount);
var inputNeurons = InputNeurons.ToArray();
var outputNeurons = OutputNeurons.ToArray();
for (int i = 0; i < InputCount; i++)
{
for (int j = 0; j < OutputCount; j++)
{
Gene newGene = new Gene(
inputNeurons[i],
outputNeurons[j],
Random.Range(-geneWeightRandomVal, geneWeightRandomVal)
);
Genes.Add(newGene);
inputNeurons[i].OutGenes.Add(newGene);
outputNeurons[j].InGenes.Add(newGene);
}
}
}
private void Initialize(int inputs, int outputs)
{
var random = new Random();
for (int i = 0; i < inputs; i++)
{
InputNeurons.Add(new Neuron());
}
for (int i = 0; i < outputs; i++)
{
var dentrites = new List <Synapse>();
foreach (var input in InputNeurons)
{
dentrites.Add(new Synapse()
{
Neuron = input, Weight = (random.NextDouble() - 0.5) * 0.01
});
}
OutputNeurons.Add(new Neuron()
{
Dendrites = dentrites
});
}
}
public double[] GetOutputValues(int cycle)
{
return(OutputNeurons.Select(x => x.Values[cycle] / Divisor).ToArray());
}
public void RemoveOutputNeuron(WorkingNeuron neuron)
{
OutputNeurons.Remove(neuron);
}
public void AddOutputNeuron(Neuron neuron)
{
OutputNeurons.Add(neuron);
}
