/// <summary>
/// Create a new child genome from a given parent genome.
/// </summary>
/// <param name="parent">The parent genome.</param>
/// <param name="rng">Random source.</param>
/// <returns>A new child genome.</returns>
public NeatGenome <T> CreateChildGenome(NeatGenome <T> parent, IRandomSource rng)
{
Debug.Assert(_metaNeatGenome == parent.MetaNeatGenome, "Parent genome has unexpected MetaNeatGenome.");
// Attempt to find a new connection that we can add to the genome.
if (!TryGetConnection(parent, rng, out DirectedConnection directedConn, out int insertIdx))
{ // Failed to find a new connection.
return(null);
}
// Determine the connection weight.
// 50% of the time use weights very close to zero.
// Note. this recreates the strategy used in SharpNEAT 2.x.
// ENHANCEMENT: Reconsider the distribution of new weights and if there are better approaches (distributions) we could use.
T weight = rng.NextBool() ? _weightSamplerB.Sample(rng) : _weightSamplerA.Sample(rng);
// Create a new connection gene array that consists of the parent connection genes plus the new gene
// inserted at the correct (sorted) position.
var parentConnArr = parent.ConnectionGenes._connArr;
var parentWeightArr = parent.ConnectionGenes._weightArr;
int parentLen = parentConnArr.Length;
// Create the child genome's ConnectionGenes object.
int childLen = parentLen + 1;
var connGenes = new ConnectionGenes <T>(childLen);
var connArr = connGenes._connArr;
var weightArr = connGenes._weightArr;
// Copy genes up to insertIdx.
Array.Copy(parentConnArr, connArr, insertIdx);
Array.Copy(parentWeightArr, weightArr, insertIdx);
// Copy the new genome into its insertion point.
connArr[insertIdx] = new DirectedConnection(
directedConn.SourceId,
directedConn.TargetId);
weightArr[insertIdx] = weight;
// Copy remaining genes (if any).
Array.Copy(parentConnArr, insertIdx, connArr, insertIdx + 1, parentLen - insertIdx);
Array.Copy(parentWeightArr, insertIdx, weightArr, insertIdx + 1, parentLen - insertIdx);
// Create and return a new genome.
// Notes.
// The set of hidden node IDs remains unchanged from the parent, therefore we are able to re-use parent.HiddenNodeIdArray.
// However, the presence of a new connection invalidates parent.NodeIndexByIdMap for use in the new genome, because the allocated
// node indexes are dependent on node depth in the acyclic graph, which in turn can be modified by the presence of a new connection.
return(_genomeBuilder.Create(
_genomeIdSeq.Next(),
_generationSeq.Peek,
connGenes,
parent.HiddenNodeIdArray));
}
/// <summary>
/// Create a new child genome from a given parent genome.
/// </summary>
/// <param name="parent">The parent genome.</param>
/// <returns>A new child genome.</returns>
public NeatGenome <T> CreateChildGenome(NeatGenome <T> parent)
{
// Attempt to find a new connection that we can add to the genome.
if (!TryGetConnection(parent, out DirectedConnection directedConn, out int insertIdx))
{ // Failed to find a new connection.
return(null);
}
// Determine the connection weight.
// 50% of the time use weights very close to zero.
// Note. this recreates the strategy used in SharpNEAT 2.x.
// TODO: Reconsider the distribution of new weights and if there are better approaches (distributions) we could use.
T weight = _rng.NextBool() ? _weightDistB.Sample() : _weightDistA.Sample();
// Create a new connection gene array that consists of the parent connection genes plus the new gene
// inserted at the correct (sorted) position.
var parentConnArr = parent.ConnectionGenes._connArr;
var parentWeightArr = parent.ConnectionGenes._weightArr;
int parentLen = parentConnArr.Length;
// Create the child genome's ConnectionGenes object.
int childLen = parentLen + 1;
var connGenes = new ConnectionGenes <T>(childLen);
var connArr = connGenes._connArr;
var weightArr = connGenes._weightArr;
// Copy genes up to insertIdx.
Array.Copy(parentConnArr, connArr, insertIdx);
Array.Copy(parentWeightArr, weightArr, insertIdx);
// Copy the new genome into its insertion point.
connArr[insertIdx] = new DirectedConnection(
directedConn.SourceId,
directedConn.TargetId);
weightArr[insertIdx] = weight;
// Copy remaining genes (if any).
Array.Copy(parentConnArr, insertIdx, connArr, insertIdx + 1, parentLen - insertIdx);
Array.Copy(parentWeightArr, insertIdx, weightArr, insertIdx + 1, parentLen - insertIdx);
// Create and return a new genome.
// Note. The set of hidden node IDs remains unchanged from the parent, therefore we are able to re-use
// both parent.HiddenNodeIdArray and NodeIndexByIdMap.
return(_genomeBuilder.Create(
_genomeIdSeq.Next(),
_generationSeq.Peek,
connGenes,
parent.HiddenNodeIdArray,
parent.NodeIndexByIdMap));
}
/// <summary>
/// Create a new child genome from a given parent genome.
/// </summary>
/// <param name="parent">The parent genome.</param>
/// <param name="rng">Random source.</param>
/// <returns>A new child genome.</returns>
public NeatGenome <T>?CreateChildGenome(NeatGenome <T> parent, IRandomSource rng)
{
// We require at least two connections in the parent, i.e. we avoid creating genomes with
// no connections, which would be pointless.
if (parent.ConnectionGenes.Length < 2)
{
return(null);
}
// Select a gene at random to delete.
var parentConnArr = parent.ConnectionGenes._connArr;
var parentWeightArr = parent.ConnectionGenes._weightArr;
int parentLen = parentConnArr.Length;
int deleteIdx = rng.Next(parentLen);
// Create the child genome's ConnectionGenes object.
int childLen = parentLen - 1;
var connGenes = new ConnectionGenes <T>(childLen);
var connArr = connGenes._connArr;
var weightArr = connGenes._weightArr;
// Copy genes up to deleteIdx.
Array.Copy(parentConnArr, connArr, deleteIdx);
Array.Copy(parentWeightArr, weightArr, deleteIdx);
// Copy remaining genes (if any).
Array.Copy(parentConnArr, deleteIdx + 1, connArr, deleteIdx, childLen - deleteIdx);
Array.Copy(parentWeightArr, deleteIdx + 1, weightArr, deleteIdx, childLen - deleteIdx);
// Get an array of hidden node IDs.
var hiddenNodeIdArr = GetHiddenNodeIdArray(parent, deleteIdx, connArr);
// Create and return a new genome.
return(_genomeBuilder.Create(
_genomeIdSeq.Next(),
_generationSeq.Peek,
connGenes,
hiddenNodeIdArr));
}
/// <summary>
/// Creates a single randomly initialised genome.
/// </summary>
private NeatGenome <T> CreateGenome()
{
// Determine how many connections to create in the new genome, as a proportion of all possible connections
// between the input and output nodes.
int connectionCount = (int)NumericsUtils.ProbabilisticRound(_connectionDefArr.Length * _connectionsProportion, _rng);
// Ensure there is at least one connection.
connectionCount = Math.Max(1, connectionCount);
// Select a random subset of all possible connections between the input and output nodes.
int[] sampleArr = new int[connectionCount];
DiscreteDistributionUtils.SampleUniformWithoutReplacement(
_connectionDefArr.Length, sampleArr, _rng);
// Sort the samples.
// Note. This results in the neural net connections being sorted by sourceID then targetID.
Array.Sort(sampleArr);
// Create the connection gene arrays and populate them.
var connGenes = new ConnectionGenes <T>(connectionCount);
var connArr = connGenes._connArr;
var weightArr = connGenes._weightArr;
for (int i = 0; i < sampleArr.Length; i++)
{
DirectedConnection cdef = _connectionDefArr[sampleArr[i]];
connArr[i] = new DirectedConnection(
cdef.SourceId,
cdef.TargetId);
weightArr[i] = _connWeightDist.Sample(_metaNeatGenome.ConnectionWeightRange, true);
}
// Get create a new genome with a new ID, birth generation of zero.
int id = _genomeIdSeq.Next();
return(_genomeBuilder.Create(id, 0, connGenes));
}
