public static void crossover(GeneticTree current, GeneticTree other)
{
List <string> currentIndices = current._nodeIndices.Where(x => x.Length != 1).ToList();
List <string> otherIndices = other._nodeIndices.Where(x => x.Length != 1).ToList();
int currentIndexAux = Rng.Rnd.Next(currentIndices.Count);
int otherIndexAux = Rng.Rnd.Next(otherIndices.Count);
string currentIndex = currentIndices[currentIndexAux];
string otherIndex = otherIndices[otherIndexAux];
GeneticNode iterCurrent = current._root;
GeneticNode iterOther = other._root;
for (int i = 1; i < currentIndex.Length; i++)
{
int index = GeneticTree._TREE_NODES_CHAR_TO_INT[currentIndex[i]];
iterCurrent = iterCurrent._children._children[index];
}
for (int i = 1; i < otherIndex.Length; i++)
{
int index = GeneticTree._TREE_NODES_CHAR_TO_INT[otherIndex[i]];
iterOther = iterOther._children._children[index];
}
current.attachNode(iterOther, currentIndex);
other.attachNode(iterCurrent, otherIndex);
current.recalculateIndices();
other.recalculateIndices();
}
public void calculatePositions()
{
int correct = 0;
GeneticTree _tree = _population[0];
Task t = new Task(() =>
{
StreamReader sr = new StreamReader("./test/testpos");
string line;
char[] fenDelim = { ';' };
while ((line = sr.ReadLine()) != null)
{
string[] FENs = line.Split(fenDelim);
string first = FENs[0].Split(null)[0];
string second = FENs[1].Split(null)[0];
Process p = new Process();
p.StartInfo.FileName = "bestmove.exe";
p.StartInfo.Arguments = $"{_evalDepth} \"{FENs[0]}\" \"{_tree._rpn}\"";
p.StartInfo.RedirectStandardOutput = true;
p.StartInfo.CreateNoWindow = true;
p.StartInfo.UseShellExecute = false;
p.Start();
string FEN = p.StandardOutput.ReadToEnd().Trim();
if (FEN.Split(null)[0].CompareTo(second) == 0)
{
correct++;
}
}
});
t.Start();
Task.WaitAll(new Task[] { t });
_statistics._positionsSolved.Add(correct);
}
public void beginCrossover()
{
_successfulCrossover = new List <TreePair>();
_toMutate = new List <GeneticTree>();
foreach (var item in _toCrossover)
{
if (Rng.Rnd.NextDouble() < _crossoverRate)
{
GeneticTree t1 = GeneticTree.deepCopy(item._first);
GeneticTree t2 = GeneticTree.deepCopy(item._second);
crossover(t1, t2);
_successfulCrossover.Add(new TreePair(GeneticTree.deepCopy(t1), GeneticTree.deepCopy(t2)));
_swapParents(t1, t2, item._first._id, item._second._id);
_toMutate.Add(t1);
_toMutate.Add(t2);
}
else
{
_toMutate.Add(item._first);
_toMutate.Add(item._second);
}
}
}
private void _swapParents(GeneticTree first, GeneticTree second, ulong firstParentId, ulong secondParentId)
{
ulong randomSalt = Rng.randomSparseBits();
first._id = (firstParentId & _RIGHT_PARENT_MASK) | (secondParentId & _LEFT_PARENT_MASK) ^ randomSalt;
second._id = (secondParentId & _RIGHT_PARENT_MASK) | (firstParentId & _LEFT_PARENT_MASK) ^ randomSalt;
first._firstParentId = second._firstParentId = firstParentId;
first._secondParentId = second._secondParentId = secondParentId;
}
public static void mutateOpportunistic(GeneticTree current)
{
NodeStatistics stats = current.allNodesStats(current._root, 0, "", 0);
List <NodeStatsPair> opportunities = new List <NodeStatsPair>();
foreach (var item in current._nodeStatsDictionary)
{
//Debug.WriteLine("Node " + item.Key + " /// Operator variance: " + item.Value._opVariance + " Variable variance: " + item.Value._varVariance + " Avg depth: " + item.Value._avgDepth + " Avg branching: " + item.Value._avgBranching);
opportunities.Add(new NodeStatsPair(item.Key, item.Value._avgBranching + item.Value._avgDepth + item.Value._opVariance + item.Value._varVariance));
}
float max = opportunities.Max(x => x._mutationValue);
foreach (var item in opportunities)
{
item._mutationValue = max - item._mutationValue + 1; // make less branchy, depthy and variant more likely to be mutated
}
opportunities = opportunities.OrderByDescending(x => x._mutationValue).ToList();
float[] wSum = new float[opportunities.Count];
wSum[0] = opportunities[0]._mutationValue;
for (int i = 1; i < wSum.Length; i++)
{
wSum[i] = opportunities[i]._mutationValue + wSum[i - 1];
}
float sumF = opportunities.Sum(x => x._mutationValue);
double centil = Rng.Rnd.NextDouble() * sumF;
int index = 0;
for (int i = 0; i < opportunities.Count && centil < wSum[i]; i++)
{
index = i;
}
string currentIndex = opportunities[index]._node;
int newDepth = (int)current._maxDepth - currentIndex.Length + 1;
GeneticTree added = new GeneticTree((uint)newDepth, current._maxBranching);
GeneticNode tmp = current._root;
for (int i = 1; i < currentIndex.Length; i++) // traversing to find proper index
{
index = GeneticTree._TREE_NODES_CHAR_TO_INT[currentIndex[i]];
tmp = tmp._children._children[index];
}
current.attachNode(added._root, currentIndex);
current.recalculateIndices();
}
public static void mutateSwap(GeneticTree current)
{
List <string> currentIndices = current._nodeIndices.Where(x => x.Length != 1).ToList();
int currentIndexAux = Rng.Rnd.Next(currentIndices.Count);
string currentIndex = currentIndices[currentIndexAux];
List <string> otherIndices = current._nodeIndices.
Where(x => x.Length != 1).ToList();
List <string> shorter = otherIndices.Where(x => x.Length <= currentIndex.Length).Where(x => currentIndex.Substring(0, x.Length).CompareTo(x) != 0).ToList();
List <string> longer = otherIndices.Where(x => x.Length > currentIndex.Length).Where(x => x.Substring(0, currentIndex.Length).CompareTo(currentIndex) != 0).ToList();
otherIndices = shorter.Union(longer).ToList();
if (otherIndices == null || otherIndices.Count == 0)
{
return;
}
int otherIndexAux = Rng.Rnd.Next(otherIndices.Count);
string otherIndex = otherIndices[otherIndexAux];
GeneticNode tmp = current._root;
GeneticNode tmp2 = current._root;
for (int i = 1; i < currentIndex.Length; i++)
{
int index = GeneticTree._TREE_NODES_CHAR_TO_INT[currentIndex[i]];
tmp = tmp._children._children[index];
}
for (int i = 1; i < otherIndex.Length; i++)
{
int index = GeneticTree._TREE_NODES_CHAR_TO_INT[otherIndex[i]];
tmp2 = tmp2._children._children[index];
}
current.attachNode(tmp2, currentIndex);
current.attachNode(tmp, otherIndex);
current.recalculateIndices();
}
public void beginMutation()
{
_successfulMutation = new List <GeneticTree>();
foreach (var item in _toMutate)
{
item.setColor(item._nodesColor);
item.allNodesStats(item._root, 0, "", 0);
if (Rng.Rnd.NextDouble() < _mutationRate)
{
GeneticTree copy = GeneticTree.deepCopy(item);
mutateOpportunistic(copy);
_successfulMutation.Add(copy);
}
else
{
_newPopulation.Add(item);
}
}
}
public static void mutateReplace(GeneticTree current)
{
List <string> currentIndices = current._nodeIndices.Where(x => x.Length != 1).ToList();
int currentIndexAux = Rng.Rnd.Next(currentIndices.Count);
string currentIndex = currentIndices[currentIndexAux];
int newDepth = (int)current._maxDepth - currentIndex.Length + 1;
GeneticTree added = new GeneticTree((uint)newDepth, current._maxBranching);
GeneticNode tmp = current._root;
for (int i = 1; i < currentIndex.Length; i++) // traversing to find proper index
{
int index = GeneticTree._TREE_NODES_CHAR_TO_INT[currentIndex[i]];
tmp = tmp._children._children[index];
}
current.attachNode(added._root, currentIndex);
current.recalculateIndices();
}
public static GeneticTree deepCopy(GeneticTree other)
{
GeneticTree res = new GeneticTree(other._maxDepth, other._maxBranching);
res._nodesColor = other._nodesColor;
res._id = other._id;
res._nodeIndices = new List <string>(other._nodeIndices);
switch (other._root)
{
case OperatorNode on:
res._root = OperatorNode.deepCopy((OperatorNode)other._root);
break;
case VariableNode vn:
res._root = VariableNode.deepCopy((VariableNode)other._root);
break;
case ConstNode cn:
res._root = ConstNode.deepCopy((ConstNode)other._root);
break;
case MultipleNode mn:
res._root = MultipleNode.deepCopy((MultipleNode)other._root);
break;
default:
break;
}
res.getRPN();
res._nodeStatsDictionary = other._nodeStatsDictionary;
res._firstParentId = other._firstParentId;
res._secondParentId = other._secondParentId;
//res.setColor(other._nodesColor.r, other._nodesColor.g, other._nodesColor.b);
return(res);
}
public void beginSelection()
{
_newPopulation = new List <GeneticTree>();
_toCrossover = new List <TreePair>();
int elitismPassNumber = (int)(_populationSize * _elitismPercentage);
if (elitismPassNumber % 2 != 0)
{
elitismPassNumber++;
}
for (int i = 0; i < elitismPassNumber; i++)
{
_newPopulation.Add(GeneticTree.deepCopy(_population[i]));
}
switch (_selectionType)
{
case GpSelectionType.RANK:
int[] weights = Enumerable.Range(1, (int)_populationSize).Reverse().ToArray();
int[] weightsRev = weights.Reverse().ToArray();
uint[] weightSum = new uint[_populationSize];
weightSum[0] = _populationSize;
for (uint i = 1; i < weightSum.Length; i++)
{
weightSum[i] = _populationSize - i + weightSum[i - 1];
}
int sum = ((1 + (int)_populationSize)) * (int)_populationSize / 2;
for (int i = 0; i < _populationSize - elitismPassNumber; i += 2)
{
int centil = Rng.Rnd.Next(0, sum);
int resIndex = weightsRev.First(x => weightSum[x - 1] > centil) - 1;
GeneticTree first = _population[resIndex];
centil = Rng.Rnd.Next(0, sum);
resIndex = weightsRev.First(x => weightSum[x - 1] > centil) - 1;
GeneticTree second = _population[resIndex];
_toCrossover.Add(new TreePair(GeneticTree.deepCopy(first), GeneticTree.deepCopy(second)));
}
break;
case GpSelectionType.ROULETTE:
weights = Enumerable.Range(0, (int)_populationSize).ToArray();
weightsRev = weights.Reverse().ToArray();
float[] wSum = new float[(int)_populationSize];
wSum[0] = _fitnesses[0];
for (int i = 1; i < wSum.Length; i++)
{
wSum[i] = _fitnesses[i] + wSum[i - 1];
}
float sumF = _fitnesses.Sum();
for (int i = 0; i < _populationSize - elitismPassNumber; i += 2)
{
double centil = Rng.Rnd.NextDouble() * sumF;
int resIndex = weights.First(x => wSum[x] > centil);
GeneticTree first = _population[resIndex];
centil = Rng.Rnd.NextDouble() * sumF;
resIndex = weights.First(x => wSum[x] > centil);
GeneticTree second = _population[resIndex];
_toCrossover.Add(new TreePair(GeneticTree.deepCopy(first), GeneticTree.deepCopy(second)));
}
break;
case GpSelectionType.TOURNAMENT:
weights = Enumerable.Range(0, (int)_populationSize).ToArray();
var shuffled = weights.OrderBy(x => Guid.NewGuid());
for (int i = 0; i < _populationSize - elitismPassNumber; i += 2)
{
var slice = shuffled.Take((int)_tournamentSelectionSize);
var firstRandom = _population[slice.OrderByDescending(x => _fitnesses[x]).First()];
shuffled = weights.OrderBy(x => Guid.NewGuid());
slice = shuffled.Take((int)_tournamentSelectionSize);
var secondRandom = _population[slice.OrderByDescending(x => _fitnesses[x]).First()];
shuffled = weights.OrderBy(x => Guid.NewGuid());
_toCrossover.Add(new TreePair(GeneticTree.deepCopy(firstRandom), GeneticTree.deepCopy(secondRandom)));
}
break;
}
}
public TreePair(GeneticTree first, GeneticTree second)
{
_first = first;
_second = second;
}
