void Reproduce()
{
Node p1, p2;
int[] parents = new int[2];
int length;
//if (stalenessCount > 5 && AILearning.MUTATION_PERCENT < .5)
// AILearning.MUTATION_PERCENT += .05;
//else if (stalenessCount <= 5 && AILearning.MUTATION_PERCENT > .05)
// AILearning.MUTATION_PERCENT -= .05;
//if (stalenessCount > 5 && Node.CrossoverPoint >= 0.1)
// Node.CrossoverPoint -= 0.1;
//else if (stalenessCount <= 5 && Node.CrossoverPoint < .75)
// Node.CrossoverPoint += .01;
// select the x% fittest population to have an orgy; slaughter the rest
int numToRemove = ((int)(GENERATION_SIZE * (1.0 - SELECTION_PERCENT))) - (GENERATION_SIZE - nodes.Count);
if (numToRemove >= 0 && nodes.Count - numToRemove >= 10)
{
nodes.RemoveRange(0, numToRemove);
}
if (nodes[nodes.Count - 1].moves.Count != nodes[0].moves.Count)
{
nodes.RemoveAt(nodes.Count - 1);
}
length = nodes.Count;
while (nodes.Count < GENERATION_SIZE)
{
// pick two parents
MyRand.GenerateRandomList(2, length, parents);
p1 = nodes[parents[0]];
p2 = nodes[parents[1]];
//Debug.Assert(p1.moves.Count == p2.moves.Count);
// ahh, the magic of life
nodes.AddRange(Node.Reproduce(p1, p2));
}
//nodes.ForEach(n => Debug.Assert(n.moves.Count == Node.NumMoves));
for (int i = 0; i < nodes.Count; i++)
{
nodes[i].x = 1;
nodes[i].y = 0;
}
}
int GenerateMove(Node n, int expectedMove)
{
return(MyRand.rand.Next(4));
int[] moves = new int[4];
MyRand.GenerateRandomList(4, moves);
int x, y;
int move;
int index = 0;
do
{
// this is essentially our repair function for invalid paths
// this introduces more variability into the population as well
move = expectedMove >= 0 ? expectedMove : moves[index++];
//move = moves[0];
x = n.x;
y = n.y;
switch (move)
{
case 0: x++;
break;
case 1: y++;
break;
case 2: x--;
break;
case 3: y--;
break;
}
expectedMove = -1;
} while (!maze.isValidMove(x, y));
return(move);
}
