/// <summary>
/// This does a fuzzy match to find the best links it can
/// </summary>
/// <remarks>
/// The existing links passed in point to where neurons used to be. This method compares those positions with where
/// neurons are now, and tries to match up with the closest it can
///
/// NOTE: Both neurons and links could have been mutated independent of each other. This method doesn't care how
/// they got the way they are, it just goes by position
/// </remarks>
private static LinkIndexed[] BuildInternalLinksExisting_Continue(IEnumerable<INeuron> neurons, int count, NeuralLinkDNA[] existing, int maxBrainChemicals, int maxIntermediateLinks, int maxFinalLinks)
{
NeuralLinkDNA[] existingPruned = existing;
if (existing.Length > count)
{
// Prune without distributing weight (if this isn't done here, then the prune at the bottom of this method will
// artificially inflate weights with the links that this step is removing)
existingPruned = existing.OrderByDescending(o => Math.Abs(o.Weight)).Take(count).ToArray();
}
Point3D[] allPoints = neurons.Select(o => o.Position).ToArray();
#region Find closest points
// Get a unique list of points
Dictionary<Point3D, ClosestExistingResult[]> resultsByPoint = new Dictionary<Point3D, ClosestExistingResult[]>(); // can't use SortedList, because point isn't sortable (probably doesn't have IComparable)
foreach (var exist in existingPruned)
{
if (!resultsByPoint.ContainsKey(exist.From))
{
resultsByPoint.Add(exist.From, GetClosestExisting(exist.From, allPoints, maxIntermediateLinks));
}
if (!resultsByPoint.ContainsKey(exist.To))
{
resultsByPoint.Add(exist.To, GetClosestExisting(exist.To, allPoints, maxIntermediateLinks));
}
}
#endregion
List<LinkIndexed> retVal = new List<LinkIndexed>();
#region Build links
foreach (var exist in existingPruned)
{
HighestPercentResult[] links = GetHighestPercent(resultsByPoint[exist.From], resultsByPoint[exist.To], maxFinalLinks, true);
foreach (HighestPercentResult link in links)
{
double[] brainChemicals = null;
if (exist.BrainChemicalModifiers != null)
{
brainChemicals = exist.BrainChemicalModifiers.
Take(maxBrainChemicals). // if there are more, just drop them
Select(o => o).
//Select(o => o * link.Percent). // I decided not to multiply by percent. The weight is already reduced, no point in double reducing
ToArray();
}
retVal.Add(new LinkIndexed(link.From.Index, link.To.Index, exist.Weight * link.Percent, brainChemicals));
}
}
#endregion
// Exit Function
if (retVal.Count > count)
{
#region Prune
// Prune the weakest links
// Need to redistribute the lost weight (since this method divided links into smaller ones). If I don't, then over many generations,
// the links will tend toward zero
retVal = retVal.OrderByDescending(o => Math.Abs(o.Weight)).ToList();
LinkIndexed[] kept = retVal.Take(count).ToArray();
LinkIndexed[] removed = retVal.Skip(count).ToArray();
double keptSum = kept.Sum(o => Math.Abs(o.Weight));
double removedSum = removed.Sum(o => Math.Abs(o.Weight));
double ratio = keptSum / (keptSum + removedSum);
ratio = 1d / ratio;
return kept.Select(o => new LinkIndexed(o.From, o.To, o.Weight * ratio, o.BrainChemicalModifiers)).ToArray();
#endregion
}
else
{
return retVal.ToArray();
}
}
public ContainerInput(long token, INeuronContainer container, NeuronContainerType containerType, Point3D position, Quaternion orientation, double? internalRatio, Tuple<NeuronContainerType, ExternalLinkRatioCalcType, double>[] externalRatios, int brainChemicalCount, NeuralLinkDNA[] internalLinks, NeuralLinkExternalDNA[] externalLinks)
{
this.Token = token;
this.Container = container;
this.ContainerType = containerType;
this.Position = position;
this.Orientation = orientation;
this.InternalRatio = internalRatio;
this.ExternalRatios = externalRatios;
this.BrainChemicalCount = brainChemicalCount;
this.InternalLinks = internalLinks;
this.ExternalLinks = externalLinks;
}