static void Main(string[] args)
{
Console.CursorVisible = false;
Console.WriteLine("Press 1 to train new neural network");
Console.WriteLine("Press 2 start snake with hardcoded AI");
Console.WriteLine("Press 3 to load trained network");
var key = Console.ReadKey();
bool isTraining = false;
SnakeAI snakeAI = null;
if (key.KeyChar == '1')
{
isTraining = true;
new Thread(() =>
{
Thread.CurrentThread.Priority = ThreadPriority.BelowNormal;
TrainNeuralAI();
}).Start();
}
else
if (key.KeyChar == '2')
{
snakeAI = new SnakeFakeAI();
}
else
{
snakeAI = new SnakeNeuralAI()
{
Network = NeuralNetwork.Load(@"AI\TrainedNetwork")
};
}
Console.Clear();
while (true)
{
if (isTraining)
{
lock (_lockObjectFile)
{
snakeAI = SnakeNeuralAI.Clone(SuperNetwork, false);
}
}
World world = new World(worldWidth, worldHeigh, foodPoints);
Stopwatch stopwatch = new Stopwatch();
while (world.Snake.Alive)
{
stopwatch.Restart();
snakeAI.MakeDecision(world);
lock (_lockObjectConsole)
{
ConsoleRenderer.Render(world);
}
Thread.Sleep(Math.Max(0, 50 - (int)stopwatch.ElapsedMilliseconds));
if (Console.KeyAvailable)
{
Console.ReadKey();
break;
}
}
lock (_lockObjectConsole)
{
ConsoleRenderer.Render(world);
}
Thread.Sleep(2000);
}
}
public static void TrainNeuralAI()
{
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
//Continue training AI
//SuperNetwork = (ActivationNetwork)ActivationNetwork.Load(@"AI\TrainedNetwork");
int bestScore = 0;
IList <SnakeNeuralAiTrainer> students = new List <SnakeNeuralAiTrainer>();
IList <SnakeNeuralAiTrainer> bestSnakeAIs = new List <SnakeNeuralAiTrainer>();
for (long i = 0; i < learningIteration / (learningPeers + learningPeersMutants); i++)
{
students.Clear();
for (int j = 0; j < learningPeers + learningPeersMutants; j++)
{
bool learnMode = j < learningPeersMutants;
SnakeNeuralAI snakeAI;
if (bestSnakeAIs.Count == parallelNetworks)
{
snakeAI = SnakeNeuralAI.Clone(bestSnakeAIs[j % parallelNetworks].SnakeNeuralAI.Network, learnMode);
}
else
{
snakeAI = SnakeNeuralAI.Clone(SuperNetwork, learnMode);
}
students.Add(new SnakeNeuralAiTrainer(snakeAI));
}
Parallel.ForEach(students, new ParallelOptions()
{
MaxDegreeOfParallelism = 6
}, (trainer) =>
{
World world = new World(trainWorldWidth, trainWorldHeigh, foodPoints);
trainer.Train(world);
});
bestSnakeAIs = students.OrderByDescending(x => x.Score).Take(parallelNetworks).ToList();
lock (_lockObjectFile)
{
if (bestSnakeAIs.First().Score > 0) // && bestOfTheBest.Score > bestScore * 0.1)
{
SuperNetwork = bestSnakeAIs.First().SnakeNeuralAI.Network;
}
}
bestScore = Math.Max(bestSnakeAIs.Max(x => x.Score), bestScore);
SuperNetwork.Save(TrainingFileName);
lock (_lockObjectConsole)
{
Console.SetCursorPosition(0, 0);
Console.WriteLine("Current average score: " + Math.Round(bestSnakeAIs.Average(x => x.Score)) + " ");
Console.WriteLine("Current best score: " + bestSnakeAIs.Max(x => x.Score) + " ");
Console.WriteLine("Learning progress: " + Math.Round((100m / (learningIteration / (learningPeers + learningPeersMutants))) * i) + "% (" + i * (learningPeers + learningPeersMutants) + " games played in " + stopwatch.Elapsed.ToString(@"hh\:mm\:ss") + ")");
}
}
}
