public static void Play(Board b)
{
Board b2 = GoDiePointers.DeepClone(b);
Player PW = new Player(true); NeuralNet NNW = new NeuralNet(PW, 3, 10);
Data.ReadNs(NNW);
Player PB = new Player(false); NeuralNet NNB = new NeuralNet(PB, 3, 10);
Data.ReadNs(NNB);
foreach (Piece p in b2.Pieces)
{
if (p is Empty)
{
continue;
}
if (p.Player.IsW == true)
{
p.Player = PW;
}
else
{
p.Player = PB;
}
}
List <Neuron> BestNeurons = GoDiePointers.DeepClone(NNW.Neurons);
Random random = new Random();
//Amount of weights to change
int changeCount = 5;
for (int j = 0; j < changeCount; j++)
{
//For neurons
//It increases/decreases the weight by rand.next(x, y)% [normally]
//It currently is used as an input for the sigmoid (as a randomizing factor)
double randomVal = random.Next(-14, 14);
//For pieces
double pieceRVal = random.Next(1, 19) / 10.00;
int randNeuron = random.Next(0, NNW.Neurons.Count);
int randthing = random.Next(1, 2);
int X = random.Next(0, 7);
int Y = random.Next(0, 7);
try
{
if (randthing == 1)
{
if (BestNeurons[randNeuron].layer == 0)
{
NNW.Neurons[randNeuron].weights[X, Y] =
Sigmoid.sigmoid(randomVal);
}
else
{
KeyValuePair <Neuron, double> kvp = NNW.Neurons[randNeuron].layWeights.ElementAt(random.Next(0, NNW.Neurons[randNeuron].layWeights.Count));
NNW.Neurons[randNeuron].layWeights[kvp.Key] =
Sigmoid.sigmoid(randomVal);
}
}
if (randthing == 2)
{
if (BestNeurons[randNeuron].layer == 0)
{
NNB.Neurons[randNeuron].weights[X, Y] = Sigmoid.sigmoid(randomVal);
}
else
{
KeyValuePair <Neuron, double> kvp = NNB.Neurons[randNeuron].layWeights.ElementAt(random.Next(0, NNB.Neurons[randNeuron].layWeights.Count));
NNB.Neurons[randNeuron].layWeights[kvp.Key] =
Sigmoid.sigmoid(randomVal);
}
}
/*
* Disabled for now
* also, it has a 50% chance of selecting the empty squares with the current x/y randomizer
* //Changing class values?
* if (randthing == 3)
* {
* b.Pieces[X, Y].CVal = (int)(pieceRVal * (GoDiePointers.DeepClone(b.Pieces[X, Y].CVal)));
* }
* //Repeat to equalize chances of neuron vs piece
* if (randthing == 4)
* {
* b2.Pieces[Y, X].CVal = (int)(pieceRVal * (GoDiePointers.DeepClone(b.Pieces[X, Y].CVal)));
* }
*/
}
catch (Exception ex) { Console.WriteLine(ex); return; }
}
//At movecap, end playing, and write whoever had a higher score to the weight list file
int moveCap = 100;
int i = 1;
//While it has not moved too many times, and while no-one has won, play
//Run in parallel?
//Using two boards to allow for different piece cvals, unless I want to put that into the NN class?
while (i <= moveCap && !b.WWin && !b.BWin && !b2.WWin && !b2.BWin && !b.Stale && !b2.Stale)
{
if (b.WTurn)
{
b2.Pieces = NNW.Move(b, true).Pieces; Board.PrintBoard(b2); b2.WTurn = false; i++;
}
if (!b2.WTurn)
{
b.Pieces = NNB.Move(b2, false).Pieces; Board.PrintBoard(b); b.WTurn = true; i++;
}
else
{
Console.WriteLine("NN Failure"); break;
}
}
//Will need to check whether pieces read/write properly in the future
//If white won, write white's data
if (b.WWin || b2.WWin) /*Data.WritePieces(b);*/ Data {
public void initNN()
{
try
{
//Randomize the weights
Random r = new Random();
for (int i = 0; i <= depth; i++)
{
if (i == 0)
{
for (int ii = 0; ii <= count - 1; ii++)
{
//Foreach space in the l1 weight array, randomize it
double[,] temps = new double[, ]
{
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) }
};
Neuron n = new Neuron(this, temps, 0, 0);
}
}
if (i >= 1 && i <= depth - 1)
{
for (int ii = 0; ii <= count - 1; ii++)
{
Neuron n = new Neuron(this, new Dictionary <Neuron, double>(), 0, i);
n.layWeights.Clear();
foreach (Neuron neu in Neurons)
{
if (neu.layer == n.layer - 1)
{
if (!n.layWeights.ContainsKey(neu))
{
//Make a connection of random weight to each neuron with a layer n - 1 lower
n.layWeights.Add(neu, /* Weight for the neuron */ Sigmoid.sigmoid(r.Next(0, 999) / 100));
}
}
}
}
}
if (i == depth)
{
//Make the final neuron (output)
Neuron n = new Neuron(this, new Dictionary <Neuron, double>(), 0, i);
n.layWeights.Clear();
foreach (Neuron neu in Neurons)
{
if (neu.layer == n.layer - 1)
{
if (!n.layWeights.ContainsKey(neu))
{
//Make a connection of random weight to each neuron with a layer n - 1 lower
n.layWeights.Add(neu, /* Weight for the neuron */ Sigmoid.sigmoid(r.Next(0, 999) / 100));
}
}
}
//Output of the NN is this neuron
Output = n;
}
}
}
//If it fails, print the error out
catch (Exception ex) { Console.WriteLine(ex); }
}
public void initNN()
{
try
{
Random r = new Random();
for (int i = 0; i <= depth; i++)
{
if (i == 0)
{
for (int ii = 0; ii <= count - 1; ii++)
{
double[,] temps = new double[, ]
{
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) },
{ Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)), Sigmoid.sigmoid(r.Next(-9, 9)) }
};
Neuron n = new Neuron(this, temps, 0, 0);
}
}
if (i >= 1 && i <= depth - 1)
{
for (int ii = 0; ii <= count - 1; ii++)
{
Neuron n = new Neuron(this, new Dictionary <Neuron, double>(), 0, i);
n.layWeights.Clear();
foreach (Neuron neu in Neurons)
{
if (neu.layer == n.layer - 1)
{
if (!n.layWeights.ContainsKey(neu))
{
n.layWeights.Add(neu, /* Weight for the neuron */ Sigmoid.sigmoid(r.Next(0, 999) / 100));
}
}
}
}
}
if (i == depth)
{
Neuron n = new Neuron(this, new Dictionary <Neuron, double>(), 0, i);
n.layWeights.Clear();
foreach (Neuron neu in Neurons)
{
if (neu.layer == n.layer - 1)
{
if (!n.layWeights.ContainsKey(neu))
{
n.layWeights.Add(neu, /* Weight for the neuron */ Sigmoid.sigmoid(r.Next(0, 999) / 100));
}
}
}
Output = n;
}
}
}
catch { initNN(); }
}