private void loadNet_Click(object sender, EventArgs e)
{
// Open the network file
using (FileStream fstream = new FileStream(netFile, FileMode.Open))
{
// Load the network from file
qAgent = new BinaryFormatter().Deserialize(fstream) as QAgent;
// REDO: load instance to singleton *** ExperienceShared has to be annotated as [Serializable]
if (chkSharedExperience.Checked && staticExperience)
{
((DeepQLearning.DRLAgent.DeepQLearnShared)qAgent.w.agents[0].brain).Init(qAgent.w.agents[0].brain.experience);
}
if (chkSharedExperience.Checked && !staticExperience)
{
((DeepQLearning.DRLAgent.DeepQLearnSharedSingleton)qAgent.w.agents[0].brain).Init(qAgent.w.agents[0].brain.experience);
}
qAgent.Reinitialize();
}
if (workerThread == null)
{
workerThread = new Thread(new ThreadStart(BackgroundThread));
workerThread.Start();
}
displayBox.Text = "QNetwork loaded successfully";
}
static void comRunQ()
{
Console.WriteLine("Q is player..?");
byte temp3 = byte.Parse(Console.ReadLine());
Console.WriteLine("How many games?");
int temp1 = int.Parse(Console.ReadLine());
Console.WriteLine("Board size?");
int temp4 = int.Parse(Console.ReadLine());
Console.WriteLine("Which opponent?");
string command = Console.ReadLine();
StreamReader reader = new StreamReader("qdatafiles.txt");
List <String> files = new List <string>();
while (reader.Peek() != -1)
{
files.Add(reader.ReadLine());
}
reader.Close();
foreach (String s in files)
{
Console.WriteLine("Using file " + s);
QAgent qagentm = new QAgent(temp3);
qagentm.Load(s);
qagentm.TurnOffExploration();
qagentm.TurnOffLearning();
RunQ(temp3, qagentm, command, temp1, temp4);
}
}
private void Init(QAgent agent, QOption option)
{
if (Benchmark)
{
BenchmarkSave.CurrentTestID = _instance.BenchmarkID;
BenchmarkSave.Runs = _instance.BenchmarkRuns;
}
else if (Mode == QAIMode.Testing && BenchmarkID != null && !BenchmarkID.Equals(""))
{
BenchmarkSave.ModelPath = _instance.BenchmarkID;
}
else
{
BenchmarkSave.CurrentTestID = agent.AI_ID().ID;
BenchmarkSave.Runs = 1;
}
Debug.Log("Running " + BenchmarkSave.ModelPath + " in mode " + Mode);
_stopwatch = Stopwatch.StartNew();
if (Tester != null)
{
Tester.Init();
}
DontDestroyOnLoad(gameObject);
switch (Mode)
{
case QAIMode.Imitating: {
_imitation = new QImitation();
break;
}
default: {
var qlCNN = new QLearningCNN(PrioritizedSweeping, option);
_qlearning = qlCNN;
_qlearning.Reset(agent);
if (Remake)
{
_qlearning.RemakeModel(agent.GetState());
}
else
{
_qlearning.LoadModel();
}
if (VisualizeNetwork)
{
_visualizer = _qlearning.CreateVisualizer();
}
qlCNN.CNN.ValuesComputed += (data, isTraining) => { if (NetworkValuesUpdated != null)
{
NetworkValuesUpdated(data, isTraining);
}
};
break;
}
}
}
public static SARS MakeSARS(this QAgent agent, QAction move, QState state)
{
move.Invoke();
var s0 = agent.GetState();
return(new SARS(state, move, s0));
}
public static QAction ToQAction(this QAgent agent, Action a)
{
return(agent.GetType().GetMethods(BindingFlags.Public | BindingFlags.NonPublic | BindingFlags.Instance)
.Where(m => m.Name == a.Method.Name)
.Where(m => m.GetCustomAttributes(typeof(QBehavior), true).Length > 0)
.Select(m => ((QBehavior)m.GetCustomAttributes(typeof(QBehavior), true).First()).ObtainAction(agent, m.Name)).First());
}
static void TrainMadQ(byte qnumber, QAgent agent, String logname, String savename, int iterations)
{
Virus virus = new Virus(2, 5);
int wins = 0, wins2 = 0;
byte oppnumber = qnumber == 1 ? (byte)2 : (byte)1;
Agent opp = new MixedAgent(0.1, false, oppnumber);
StreamWriter writer = new StreamWriter(logname);
for (int i = 1; i <= iterations; i++)
{
int winner = RunGame(virus, qnumber == 1 ? (Agent)agent : opp, qnumber == 2 ? (Agent)agent : opp);
wins += winner == 1 ? 1 : 0;
wins2 += winner == 1 ? 1 : 0;
if (i % 100 == 0)
{
writer.WriteLine(wins2);
wins2 = 0;
}
if (i % 10000 == 0)
{
agent.Save(savename);
Console.WriteLine("Iteration: " + i);
Console.WriteLine("Wins: " + wins);
wins = 0;
}
virus = new Virus(2, 5);
}
writer.Close();
}
static void Main(string[] args)
{
QAgent <int, int> agent = new QAgent <int, int>();
int secretNumber = 5;
//for each state we can add 1 and subtract 1
agent.RequestActionsForState = (int state) => new List <int>()
{
-1, 1
};
for (int i = 0; i < 10; i++)
{
int guessedNumber = 0;
Console.WriteLine("--------------------------");
do
{
//flag if game is finished
bool finish = false;
int prevstate = guessedNumber;
var action = agent.GetAction(guessedNumber);
var reward = -0.1;
//execute action
guessedNumber += action;
Console.WriteLine(guessedNumber);
//specify the rule that guessed number cant be less than 0
if (guessedNumber < 0)
{
reward = -1;
finish = true;
}
//rule for victory
if (guessedNumber == secretNumber)
{
reward = 1;
finish = true;
}
//learn what would happen after action is executed
agent.UpdateState(prevstate, action, reward, guessedNumber);
//finish the game
if (finish)
{
break;
}
} while (true);
}
Console.ReadLine();
}
public override bool SetupNextTest(QAgent agent)
{
if (_testStarted)
{
return(false);
}
_timeStart = Time.time;
_testStarted = true;
return(true);
}
static void comTrainMadQ()
{
Console.WriteLine("Q is player..?");
byte temp3 = byte.Parse(Console.ReadLine());
QAgent qagent = new QAgent(temp3);
qagent.Load("qmad");
qagent.MinLearning = 0.05;
Console.WriteLine("How many iterations?");
int temp1 = int.Parse(Console.ReadLine());
TrainMadQ(temp3, qagent, "qmadlog", "qmad", temp1);
}
static void RunQ(byte qnumber, QAgent agent, String opponent, int iterations, int size)
{
Virus virus = new Virus(2, size);
int wins = 0;
byte oppnumber = qnumber == 1 ? (byte)2 : (byte)1;
Agent opp;
switch (opponent)
{
case "brute":
opp = new BruteForceAgent(oppnumber);
break;
case "minimax4":
opp = new MinimaxAgent(4, oppnumber);
break;
case "minimax3":
opp = new MinimaxAgent(3, oppnumber);
break;
case "minimax2":
opp = new MinimaxAgent(2, oppnumber);
break;
default:
opp = new BruteForceAgent(oppnumber);
break;
}
int n = 0;
while (File.Exists("qwinlog" + n))
{
n++;
}
StreamWriter writer = new StreamWriter("qwinlog" + n);
for (int i = 1; i <= iterations; i++)
{
int winner = RunGame(virus, qnumber == 1 ? (Agent)agent : opp, qnumber == 2 ? (Agent)agent : opp);
wins += winner == 1 ? 1 : 0;
virus = new Virus(2, size);
}
writer.WriteLine(wins);
Console.WriteLine("Iteration: " + iterations);
Console.WriteLine("Wins: " + wins);
wins = 0;
writer.Close();
}
public static void Imitate(QAgent agent, QState state, Action a)
{
if (_instance == null || _instance.Mode != QAIMode.Imitating)
{
return;
}
var terminal = _instance._imitation.Imitate(state, agent.ToQAction(a));
if (terminal)
{
// _instance._imitation.Save(); // Saving is now done in the Option Window, where the learning is started.
EditorApplication.isPlaying = false;
}
}
public override bool SetupNextTest(QAgent agent)
{
Running = true;
Time.timeScale = 3;
_scores.Add(new TestScore());
_iteration++;
FindObjectOfType <PongBall>().Reset(new Vector2(-1, y));
y += 0.2f;
if (Math.Abs(y) < 0.01) //y == 0
{
y += 0.2f;
}
return(y <= 1);
}
private void loadNet_Click(object sender, EventArgs e)
{
// Load the netwok from file
using (FileStream fstream = new FileStream(netFile, FileMode.Open))
{
qAgent = new BinaryFormatter().Deserialize(fstream) as QAgent;
qAgent.Reinitialize();
}
if (workerThread == null)
{
workerThread = new Thread(new ThreadStart(BackgroundThread));
workerThread.Start();
}
}
public override void OnActionTaken(QAgent agent, QAction action, QState state)
{
var car = (SlotCar)agent;
if (!car.OnTrack && !_isCrashing)
{
_isCrashing = true;
_crashes++;
}
else if (car.OnTrack && _isCrashing)
{
_isCrashing = false;
}
_maxDist = car.DistanceTravelled - car.StartPosition;
}
public static void InitAgent(QAgent agent, QOption option = null)
{
option = option ?? new QOption();
if (_instance == null)
{
_instance = FindObjectOfType <QAIManager>();
_instance.Init(agent, option);
}
BenchmarkSave.SaveBenchmarks = _instance.Benchmark;
_instance._sceneIsOver = false;
_instance._testIsOver = false;
Agent = agent;
if (_instance.Mode != QAIMode.Imitating)
{
_instance._qlearning.Reset(agent);
}
}
public override bool SetupNextTest(QAgent agent)
{
var visualizer = FindObjectOfType <GridResultsVisualizer>();
visualizer.enabled = true;
visualizer.DrawResults(_results);
if (_positions.Count == 0)
{
return(false);
}
RunPosistion = _positions[0];
RunPosistion.y = 1;
((GridWoman)agent).transform.position = RunPosistion;
_positions.RemoveAt(0);
return(true);
}
static void GatherQDataFull(String opponent, int iterations, int saveinterval, byte player, double disc, double lrmod, double lrstrt, double exmod, double exstrt, bool lrsqrt, bool exsqrt, double randomness, double minlearn, double initvalue = 0)
{
QAgent agent;
String name = opponent + disc + "Q_L" + lrmod + (lrsqrt ? "persqrt" : "per1") + "_E" + exmod + (exsqrt ? "persqrt" : "per1") + "r" + randomness + "m" + minlearn + "ls" + lrstrt + "xs" + exstrt + "i" + initvalue;
Console.WriteLine("Gathering data for Q: " + name + ":");
for (int i = 0; i < 10; i++)
{
if (!File.Exists(name + ".log" + i))
{
agent = new QAgent(player, disc, lrmod, lrstrt, exmod, exstrt, lrsqrt ? 0.5 : 1, exsqrt ? 0.5 : 1, initvalue);
agent.RandomRate = randomness;
agent.MinLearning = minlearn;
TrainQ(5, player, agent, opponent, name + ".log" + i, name + ".sav" + i, iterations, saveinterval);
}
}
}
public QAction ObtainAction(QAgent agent, string name)
{
if (_conditional)
{
var p =
agent.GetType()
.GetMethods(BindingFlags.Public | BindingFlags.NonPublic | BindingFlags.Instance)
.FirstOrDefault(m => m.Name == _predicate);
if (p == default(MethodInfo))
{
throw new Exception("Predicate method " + _predicate + " does not exist.");
}
if (p.GetCustomAttributes(typeof(QPredicate), true).Length == 0)
{
throw new Exception("Predicate method " + _predicate + " is not properly annotated as a QPredicate.");
}
}
return(new QAction(
name,
(Action)Delegate.CreateDelegate(typeof(Action), agent, name),
_conditional ? (QPredicate.Basic)Delegate.CreateDelegate(typeof(QPredicate.Basic), agent, _predicate) : null
));
}
public TTTModule() : base("/ttt")
{
QAgent <string, int> agentX = new QAgent <string, int>();
agentX.RequestActionsForState = SpaceIndexes;
QAgent <string, int> agentO = new QAgent <string, int>();
agentO.RequestActionsForState = SpaceIndexes;
Post("/predict", o =>
{
var name = this.Request.Query["name"].ToString();
var reader = new StreamReader(Request.Body);
var state = reader.ReadToEnd();
if (File.Exists($"agent{name}.json"))
{
agentX.Deserialize(File.ReadAllText($"agent{name}.json"));
}
return(agentX.GetAction(state).ToString());
});
Post("/learn", o =>
{
var name = this.Request.Query["name"].ToString();
if (File.Exists($"agent{name}.json"))
{
agentX.Deserialize(File.ReadAllText($"agent{name}.json"));
}
agentX.UpdateState(this.Request.Form["prevState"].ToString(), int.Parse(this.Request.Form["action"].ToString()), int.Parse(this.Request.Form["reward"].ToString()), this.Request.Form["newState"].ToString());
File.WriteAllText($"agent{name}.json", agentX.Serialize());
return("OK");
});
}
static void comTrainQ(bool load)
{
Console.WriteLine("Q is player..?");
byte temp3 = byte.Parse(Console.ReadLine());
QAgent qagent = new QAgent(temp3);
if (load)
{
qagent.Load("TrainingData");
}
Console.WriteLine("How many iterations?");
int temp1 = int.Parse(Console.ReadLine());
Console.WriteLine("Save how often?");
int temp2 = int.Parse(Console.ReadLine());
Console.WriteLine("Board size?");
int temp4 = int.Parse(Console.ReadLine());
Console.WriteLine("Which opponent?");
string command = Console.ReadLine();
TrainQ(temp4, temp3, qagent, command, "log", "TrainingData", temp1, temp2);
}
public abstract bool SetupNextTest(QAgent agent);
public abstract void OnActionTaken(QAgent agent, QAction action, QState state);
static void Main(string[] args) // b r o k e n
{
Console.ForegroundColor = ConsoleColor.DarkMagenta;
if (File.Exists(qAgentBrainPath))
{
using (FileStream fstream = new FileStream(qAgentBrainPath, FileMode.Open, FileAccess.Read, FileShare.Read)) {
qAgent = new BinaryFormatter().Deserialize(fstream) as QAgent;
qAgent.Reinitialize();
}
Console.WriteLine("QAgent loaded");
}
else
{
var num_inputs = 6; // 9 eyes, each sees 3 numbers (wall, green, red thing proximity)
var num_actions = 3; // 5 possible angles agent can turn
var temporal_window = 1; // amount of temporal memory. 0 = agent lives in-the-moment :)
var network_size = num_inputs * temporal_window + num_actions * temporal_window + num_inputs;
// config brain
var layer_defs = new List <LayerDefinition>();
// the value function network computes a value of taking any of the possible actions
// given an input state. Here we specify one explicitly the hard way
// but user could also equivalently instead use opt.hidden_layer_sizes = [20,20]
// to just insert simple relu hidden layers.
layer_defs.Add(new LayerDefinition {
type = "input", out_sx = 1, out_sy = 1, out_depth = network_size
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "regression", num_neurons = num_actions
});
// options for the Temporal Difference learner that trains the above net
// by backpropping the temporal difference learning rule.
//var opt = new Options { method="sgd", learning_rate=0.01, l2_decay=0.001, momentum=0.9, batch_size=10, l1_decay=0.001 };
Options opt = new Options {
method = "adadelta", l2_decay = 0.001, batch_size = 10
};
TrainingOptions tdtrainer_options = new TrainingOptions();
tdtrainer_options.temporal_window = temporal_window;
tdtrainer_options.experience_size = 30000;
tdtrainer_options.start_learn_threshold = 1000;
tdtrainer_options.gamma = 0.7;
tdtrainer_options.learning_steps_total = 200000;
tdtrainer_options.learning_steps_burnin = 3000;
tdtrainer_options.epsilon_min = 0.05;
tdtrainer_options.epsilon_test_time = 0.00;
tdtrainer_options.layer_defs = layer_defs;
tdtrainer_options.options = opt;
DeepQLearn brain = new DeepQLearn(num_inputs, num_actions, tdtrainer_options);
qAgent = new QAgent(brain);
}
qAgent.startlearn();
new Thread(() => {
while (true)
{
if (DateTime.Now.Ticks / TimeSpan.TicksPerMillisecond % 31 /*arbitrary*/ == 0)
{
using (FileStream fstream = new FileStream(qAgentBrainPath, FileMode.Create, FileAccess.Write, FileShare.ReadWrite)) {
new BinaryFormatter().Serialize(fstream, qAgent);
}
}
qAgent.tick();
}
}).Start();
}
public void Reset(QAgent agent)
{
Actions = agent.GetQActions().AsReadOnly();
}
public LabyrinthModule() : base("/labyrinth")
{
QAgent <int, string> agent = new QAgent <int, string>();
agent.RequestActionsForState = i =>
{
var res = new List <string>()
{
"up", "down", "left", "right"
};
int y = i / 10;
int x = i - y * 10;
if (x == 0)
{
res.Remove("left");
}
if (x == 4)
{
res.Remove("right");
}
if (y == 4)
{
res.Remove("down");
}
if (y == 0)
{
res.Remove("up");
}
return(res);
};
Post("/predict", o =>
{
var name = this.Request.Query["name"].ToString();
var reader = new StreamReader(Request.Body);
var state = reader.ReadToEnd();
if (File.Exists($"agent{name}.json"))
{
agent.Deserialize(File.ReadAllText($"agent{name}.json"));
}
return(agent.GetAction(int.Parse(state)).ToString());
});
Post("/learn", o =>
{
var name = this.Request.Query["name"].ToString();
if (File.Exists($"agent{name}.json"))
{
agent.Deserialize(File.ReadAllText($"agent{name}.json"));
}
agent.UpdateState(int.Parse(this.Request.Form["prevState"].ToString()), this.Request.Form["action"].ToString(), int.Parse(this.Request.Form["reward"].ToString()), int.Parse(this.Request.Form["newState"].ToString()));
File.WriteAllText($"agent{name}.json", agent.Serialize());
return("OK");
});
Post("/overlay", o =>
{
var name = this.Request.Query["name"].ToString();
if (File.Exists($"agent{name}.json"))
{
agent.Deserialize(File.ReadAllText($"agent{name}.json"));
}
double[][] res = new double[5][];
for (int y = 0; y < 5; y++)
{
res[y] = new double[5];
for (int x = 0; x < 5; x++)
{
var state = y * 10 + x;
res[y][x] = agent.GetStateMaxScore(state);
}
}
return(JsonConvert.SerializeObject(res));
});
}
public VirusInterface(Virus virus, int tilesize = 20, bool immediateAI = false, params String[] names)
{
InitializeComponent();
this.virus = virus;
this.tileSize = tilesize;
this.immediateAI = immediateAI;
this.MouseClick += MouseClickHandler1;
this.Size = new Size(
virus.Size * tileSize + 17,
virus.Size * tileSize + 55);
this.names.Add("Player 0");
this.names.AddRange(names);
while (this.names.Count < virus.Players + 1)
{
this.names.Add("Player " + this.names.Count);
}
//Save("Lalalafil");
agents = new Agent[this.names.Count];
int n = 1;
for (byte i = 1; i < this.names.Count; i++)
{
String p = this.names[i];
switch (p)
{
case "QAI":
agents[i] = new QAgent(i);
if (File.Exists("TrainingData.Q") && File.Exists("TrainingData.N"))
{
((QAgent)agents[i]).Load("TrainingData");
((QAgent)agents[i]).TurnOffExploration();
((QAgent)agents[i]).TurnOffLearning();
}
this.names[i] = "AI " + n;
n++;
break;
case "AnnAI":
agents[i] = new AnnAgent(false, virus.Size, i);
this.names[i] = "AI " + n;
n++;
break;
case "MinimaxAI":
agents[i] = new MinimaxAgent(4, i);
this.names[i] = "AI " + n;
n++;
break;
case "MiniMaxMixAI":
if (File.Exists("TrainingData.Q"))
{
agents[i] = new MiniMaxMixAgent("TrainingData", 2, i);
}
else
{
agents[i] = new BruteForceAgent(i);
}
this.names[i] = "AI " + n;
n++;
break;
case "MixedAI":
agents[i] = new MixedAgent(0.5, false, i);
this.names[i] = "AI " + n;
n++;
break;
case "BruteAI":
agents[i] = new BruteForceAgent(i);
this.names[i] = "AI " + n;
n++;
break;
case "RandomAI":
agents[i] = new RandomAgent(i);
this.names[i] = "AI " + n;
n++;
break;
case "SimpleAI":
agents[i] = new SimpleAgent(i);
this.names[i] = "AI " + n;
n++;
break;
}
}
message = this.names[1] + "'s turn";
colors = new Color[virus.Players + 1];
colors[0] = Color.White;
colors[1] = Color.FromArgb(128, 160, 255);
colors[2] = Color.FromArgb(96, 255, 96);
if (virus.Players > 2)
{
colors[3] = Color.FromArgb(255, 96, 96);
}
if (virus.Players > 3)
{
colors[4] = Color.FromArgb(255, 255, 64);
}
Random rand = new Random();
for (int i = 5; i <= virus.Players; i++)
{
colors[i] = Color.FromArgb(rand.Next(256), rand.Next(256), rand.Next(256));
}
}
public override void OnActionTaken(QAgent agent, QAction action, QState state)
{
var distToGoal = (((GridWoman)agent).transform.position - Goal.Position).magnitude;
_distScores.Add(1 / (distToGoal + 1));
}
public override void OnActionTaken(QAgent agent, QAction action, QState state)
{
}
private void startLearning_Click(object sender, EventArgs e)
{
if (qAgent == null)
{
var num_inputs = 27; // 9 eyes, each sees 3 numbers (wall, green, red thing proximity)
var num_actions = 5; // 5 possible angles agent can turn
var temporal_window = 4; // amount of temporal memory. 0 = agent lives in-the-moment :)
var network_size = num_inputs * temporal_window + num_actions * temporal_window + num_inputs;
var layer_defs = new List <LayerDefinition>();
// the value function network computes a value of taking any of the possible actions
// given an input state. Here we specify one explicitly the hard way
// but user could also equivalently instead use opt.hidden_layer_sizes = [20,20]
// to just insert simple relu hidden layers.
layer_defs.Add(new LayerDefinition {
type = "input", out_sx = 1, out_sy = 1, out_depth = network_size
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "regression", num_neurons = num_actions
});
// options for the Temporal Difference learner that trains the above net
// by backpropping the temporal difference learning rule.
//var opt = new Options { method="sgd", learning_rate=0.01, l2_decay=0.001, momentum=0.9, batch_size=10, l1_decay=0.001 };
var opt = new Options {
method = "adadelta", l2_decay = 0.001, batch_size = 10
};
var tdtrainer_options = new TrainingOptions();
tdtrainer_options.temporal_window = temporal_window;
tdtrainer_options.experience_size = 30000;
tdtrainer_options.start_learn_threshold = 1000;
tdtrainer_options.gamma = 0.7;
tdtrainer_options.learning_steps_total = 200000;
tdtrainer_options.learning_steps_burnin = 3000;
tdtrainer_options.epsilon_min = 0.05;
tdtrainer_options.epsilon_test_time = 0.00;
tdtrainer_options.layer_defs = layer_defs;
tdtrainer_options.options = opt;
var brain = new DeepQLearn(num_inputs, num_actions, tdtrainer_options);
qAgent = new QAgent(brain, canvas.Width, canvas.Height);
}
else
{
qAgent.startlearn();
}
if (workerThread == null)
{
workerThread = new Thread(new ThreadStart(BackgroundThread));
workerThread.Start();
}
}
static void TrainQ(int size, byte qnumber, QAgent agent, String opponent, String logname, String savename, int iterations, int saveinterval = 360)
{
Virus virus = new Virus(2, size);
int wins = 0, wins2 = 0;
byte oppnumber = qnumber == 1 ? (byte)2 : (byte)1;
Agent opp = new BruteForceAgent(oppnumber);
StreamWriter writer = new StreamWriter(logname);
for (int i = 1; i <= iterations; i++)
{
switch (opponent)
{
case "brute":
break;
case "minimax4":
opp = new MinimaxAgent(4, oppnumber);
break;
case "minimax3":
opp = new MinimaxAgent(3, oppnumber);
break;
case "minimax2":
opp = new MinimaxAgent(2, oppnumber);
break;
default:
opp = new BruteForceAgent(oppnumber);
break;
}
int winner = RunGame(virus, qnumber == 1 ? (Agent)agent : opp, qnumber == 2 ? (Agent)agent : opp);
wins += winner == 1 ? 1 : 0;
wins2 += winner == 1 ? 1 : 0;
if (i % 100 == 0)
{
if (agent.RandomRate == 0)
{
writer.WriteLine(wins2);
}
wins2 = 0;
}
if (i % saveinterval == 0)
{
agent.Save(savename);
Console.WriteLine("Iteration: " + i);
Console.WriteLine("Wins: " + wins);
wins = 0;
if (agent.RandomRate > 0)
{
agent.TurnOffExploration();
agent.TurnOffLearning();
for (int j = 1; j <= 1000; j++)
{
virus = new Virus(2, size);
winner = RunGame(virus, qnumber == 1 ? (Agent)agent : opp, qnumber == 2 ? (Agent)agent : opp);
wins += winner == 1 ? 1 : 0;
}
writer.WriteLine(wins);
wins = 0;
agent.TurnOnExploration();
agent.TurnOnLearning();
}
}
virus = new Virus(2, size);
}
writer.Close();
}