public MarkovChain(MarkovChainNode[] nodes, int stepsPerActivation, FastRandom random)
{
_nodes = nodes;
_stepsPerActivation = stepsPerActivation;
_random = random;
_rouletteWheels = new RouletteWheelLayout[nodes.Length];
for (int i = 0; i < nodes.Length; i++)
_rouletteWheels[i] = new RouletteWheelLayout(nodes[i].TransitionProbabilities);
}
public MarkovChain Decode(NeatGenome genome)
{
Debug.Assert(genome.InputNodeCount == 1);
MarkovChainNode[] nodes = new MarkovChainNode[genome.NeuronGeneList.Count];
List <int>[] dest = new List <int> [nodes.Length];
List <double>[] probs = new List <double> [nodes.Length];
Dictionary <uint, int> geneToNode = new Dictionary <uint, int>();
for (int i = 0; i < nodes.Length; i++)
{
nodes[i] = new MarkovChainNode()
{
State = getState(genome.NeuronGeneList[i])
};
geneToNode[genome.NeuronGeneList[i].Id] = i;
}
// Do not permit the nodes to go backwards to the non-text generating states
uint originId = genome.NeuronGeneList[0].Id;
uint biasId = genome.NeuronGeneList[1].Id;
for (int i = 0; i < nodes.Length; i++)
{
dest[i] = new List <int>();
probs[i] = new List <double>();
if (genome.NeuronGeneList[i].NodeType == NodeType.Bias)
{
continue;
}
if (_scheme.RelaxingActivation && genome.NeuronGeneList[i].NodeType == NodeType.Output)
{
continue;
}
// Find all connections with this node at the start,
// filtering out any that may lead to invalid nodes like the
// bias or the origin.
var conns = genome.ConnectionList.Where(t => t.SourceNodeId == genome.NeuronGeneList[i].Id &&
t.TargetNodeId != originId &&
t.TargetNodeId != biasId
);
if (conns == null)
{
continue;
}
double sum = conns.Sum(c => Math.Abs(c.Weight));
foreach (var conn in conns)
{
dest[i].Add(geneToNode[conn.TargetNodeId]);
probs[i].Add(Math.Abs(conn.Weight) / sum);
}
}
for (int i = 0; i < nodes.Length; i++)
{
nodes[i].TransitionDestinations = dest[i].ToArray();
nodes[i].TransitionProbabilities = probs[i].ToArray();
}
if (_scheme.RelaxingActivation)
{
return(new MarkovChain(nodes, _scheme.MaxTimesteps));
}
else
{
return(new MarkovChain(nodes, _scheme.TimestepsPerActivation));
}
}
public MarkovChain(MarkovChainNode[] nodes, int stepsPerActivation)
: this(nodes, stepsPerActivation, new FastRandom())
{
}
