static void Step4()
{
Console.WriteLine("STEP 4: Create neural network...");
NetworkCreator networkCreator = new NetworkCreator();
networkCreator.Create(DataFilesInfoGetter.NetworkFile);
}
public void TestCategories()
{
var validationSet = new List <TrainingSet>();
var trainingSet = new List <TrainingSet>();
var root = new Category("document");
var doc_1040 = root.AddSubcategory("1040");
var doc_990 = root.AddSubcategory("990");
var doc_1040_2012 = doc_1040.AddSubcategory("2012");
var doc_1040_2010 = doc_1040.AddSubcategory("2010");
var doc_1040_other = doc_1040.AddSubcategory("other");
validationSet.Add(new TrainingSet("trainingdata/documents/1040/other/tax1040.gif", doc_1040_other));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/other/tax1040_2.png", doc_1040_other));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/2012/tax1040_3.gif", doc_1040_2012));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/2010/tax1040_4.jpg", doc_1040_2010));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/other/tax1040_5.gif", doc_1040_other));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/other/tax1040_6.gif", doc_1040_other));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/other/tax1040_7.gif", doc_1040_other));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/other/tax1040_8.gif", doc_1040_other));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/2010/tax1040_9.jpg", doc_1040_2010));
validationSet.Add(new TrainingSet("trainingdata/documents/1040/2012/tax1040_10.gif", doc_1040_2012));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990.jpg", doc_990));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990_2.jpg", doc_990));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990_3.jpg", doc_990));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990_4.jpg", doc_990));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990_5.jpg", doc_990));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990_6.jpg", doc_990));
validationSet.Add(new TrainingSet("trainingdata/documents/990/tax990_7.jpg", doc_990));
var t = new Trainer();
var nc = new NetworkCreator();
var nl = nc.CreateNetworks(root);
t.Train(nl, validationSet);
var o = t.Run(nl, root, "trainingdata/documents/990/tax990_poor.gif");
o = t.Run(nl, root, "trainingdata/documents/1040/2012/tax1040_10.gif");
o = t.Run(nl, root, "trainingdata/documents/1040/2010/tax1040_4.jpg");
var x = o;
}
public void TestNetwork_FromDirectory()
{
var validationSet = TrainingSet.FromDirectory("trainingdata");
var trainingSet = TrainingSet.FromDirectory("generateddata");
var t = new Trainer();
var nc = new NetworkCreator();
var nl = nc.CreateNetworks(trainingSet.Item2);
t.Train(nl, trainingSet.Item1, validationSet.Item1);
var o = t.Run(nl, trainingSet.Item2, "trainingdata/documents/990/tax990_poor.gif");
o = t.Run(nl, trainingSet.Item2, "trainingdata/documents/1040/2012/tax1040_10.gif");
o = t.Run(nl, trainingSet.Item2, "trainingdata/documents/1040/2010/tax1040_4.jpg");
var x = o;
}
//Generate the classification tree and start training it recursively
public void RunNetwork()
{
var validationSet = TrainingSet.FromDirectory(path + "/trainingdata");
var trainingSet = TrainingSet.FromDirectory(path + "/generateddata");
root = trainingSet.Item2;
Invalidate();
//Invoke(new MethodInvoker(Invalidate));
var t = new Trainer();
var nc = new NetworkCreator();
var nl = nc.CreateNetworks(root);
t.Train(nl, trainingSet.Item1, validationSet.Item1);
new Thread(() =>
{
var thisImagePath = path + "/trainingdata/documents/990/tax990.jpg";
results = t.Run(nl, trainingSet.Item2, thisImagePath);
Invalidate();
var iv = new ImageView(Image.FromFile(thisImagePath));
iv.Visible = false;
iv.ShowDialog();
thisImagePath = path + "/trainingdata/documents/1040/2012/tax1040_10.gif";
results = t.Run(nl, trainingSet.Item2, thisImagePath);
Invalidate();
iv = new ImageView(Image.FromFile(thisImagePath));
iv.Visible = false;
iv.ShowDialog();
thisImagePath = path + "/trainingdata/documents/1040/2010/tax1040_4.jpg";
results = t.Run(nl, trainingSet.Item2, thisImagePath);
Invalidate();
iv = new ImageView(Image.FromFile(thisImagePath));
iv.Visible = false;
iv.ShowDialog();
}).Start();
}
public static void TestBackpropagation()
{
var NN = NetworkCreator.Perceptron(2, 1, new [] { 20, 20, 20 });
var nums = 100000;
var data = new Dictionary <List <double>, List <double> >();
var rand = new Random();
var dif = 1000;
var shell = 500;
for (var i = 0; i < nums; i++)
{
var first = rand.Next(dif) - shell;
var second = rand.Next(dif) - shell;
data.Add(new List <double>()
{
first, second
},
new List <double>()
{
(first + second + 2 * shell) / 2d / dif,
});
}
BackpropagationAlgorithm.Teach(NN, data);
NN.Print();
for (var i = 0; i < 100; i++)
{
var first = rand.Next(dif) - shell;
var second = rand.Next(dif) - shell;
NN.Work(new List <double>()
{
first, second
});
var res = NN.GetResults()[0];
Console.WriteLine(first + " " + second + " . " + (first + second) + " " + (Math.Round((res * dif - shell) * 2)));
Console.ReadLine();
}
}
public static void TestGeneticAlgorithm()
{
var agents = new List <NeuralEnvironment>();
var population = 100;
var rand = new Random();
var dileme = new List <double>();
var options = 7;
for (var i = 0; i < options * options * 2; i++)
{
dileme.Add(rand.Next(11) / 10d);
}
var bacldileme = new List <double>();
for (var i = 0; i < options * options; i++)
{
bacldileme.Add(dileme[i * 2 + 1]);
bacldileme.Add(dileme[i * 2]);
}
for (var i = 0; i < options; i++)
{
for (var j = 0; j < options; j++)
{
Console.Write(Math.Round(dileme[2 * (i * options + j)] * 10)
+ " " + Math.Round(dileme[2 * (i * options + j) + 1] * 10));
if (j + 1 != options)
{
Console.Write(" | ");
}
}
Console.WriteLine();
}
Console.WriteLine();
Console.WriteLine();
for (var i = 0; i < population; i++)
{
agents.Add(NetworkCreator.Perceptron(8, options, new List <int>()
{
20
}));
}
var tournament = new Func <bool, Func <List <NeuralEnvironment>, List <double> > >(print => ((players) =>
{
var startElo = 0d;
var tours = 100;
var elo =
players.ToDictionary(player => player, player => startElo);
for (var game = 0; game < tours; game++)
{
//if(print)Console.WriteLine(game);
var pairing = new List <(NeuralEnvironment, NeuralEnvironment)>();
var list = players.OrderBy(p => - elo[p]).ToList();
for (var i = 0; i < players.Count / 2; i++)
{
var f = list[rand.Next(list.Count)];
list.Remove(f);
var s = list[rand.Next(list.Count)];
list.Remove(s);
pairing.Add((f, s));
}
Parallel.ForEach(pairing, pair =>
//foreach(var pair in pairing)
{
var(player1, player2) = pair;
player1.Work(dileme);
player2.Work(bacldileme);
var firstPlay = player1.GetMaxOutId();
var secondPlay = player2.GetMaxOutId();
//winwin
elo[player1] += dileme[firstPlay * 6 + secondPlay * 2];
elo[player2] += dileme[firstPlay * 6 + secondPlay * 2 + 1];
/*var E1 = 1 / (1 + Math.Pow(10, (elo[player2] - elo[player1]) / 400));
* var E2 = 1 / (1 + Math.Pow(10, (elo[player1] - elo[player2]) / 400));
* elo[player1] += 20 * (res[1] / points + (res[0] / draw) / 2d - E1);
* elo[player2] += 20 * (res[2] / points + (res[0] / draw) / 2d- E2);*/
//}
});
}
var bestPlayer = elo.OrderBy(p => p.Value).Last().Key;
if (print)
{
Console.WriteLine(Math.Round(elo[bestPlayer])
+ " "
+ Math.Round(elo.Select(p => p.Value).Sum() * 1d / elo.Count));
}
return(players.Select(p => elo[p]).ToList());
}));
var iterations = 1000;
for (var i = 0; i < iterations; i++)
{
agents = GeneticAlgorithm.Improve(
agents,
GeneticAlgorithm.RandomMerge,
tournament(false));
}
foreach (var agent in agents.Take(10))
{
agent.Work(dileme);
Console.Write(agent.GetMaxOutId() + " ");
agent.Work(bacldileme);
Console.WriteLine(agent.GetMaxOutId());
}
}
static void Main(string[] args)
{
// All in-game parameters can be changed in GameRules.cs (but may cause some exceptions)
// Example of using library and making / teaching bot to play a game
// NetworkCreator.Perceptron(inputs, outputs, list of neurons in hidden layers) - to create agent
// NetworkCreator.ReadFromFile(filename) - to load agent from file
// PlayGame(agents) - to watch agents in action
// Teach(agents, number of example games, number of learning cycles) - to teach agents with backpropagation algorithm
// Gen(agents, number of generations) - to improve agents with genetic algorithm
// Shows perfect average lifetime for agents in game with current values
_actions = new List <int>();
for (var i = 0; i < GameRules.NumberOfActions; i++)
{
_actions.Add(0);
}
Console.WriteLine("Average target = " + (GameRules.StartOil
+ GameRules.OilProb
* GameRules.OilBonus
* GameRules.MapWidth
* GameRules.MapHeight
/ GameRules.numberOfAgents));
Console.WriteLine();
Console.WriteLine();
///////////////////////////////////////////////////////////////////////////////////////////////////////////
// Code Examples
/*
* //Creates new agents
*
* var inputs = ((GameRules.SightWidth * 2 + 1) * (GameRules.SightHeight) - 1) * GameRules.Tags.Count + 2;
* var outputs = GameRules.NumberOfActions;
* var neuronsInHiddenLayers = new List<int>() {20};
* var agents = new List<NeuralEnvironment>();
* for(var i = 0; i < GameRules.numberOfAgents; i++)
* agents.Add(NetworkCreator.Perceptron(inputs, outputs, neuronsInHiddenLayers ));
*/
/*
* //Get saved agents from files
*
* var filename = "agent";
* var iter = GameRules.numberOfAgents;
* var agents = new List<NeuralEnvironment>();
* for(var i = 0; i < iter; i++)
* agents.Add(NetworkCreator.ReadFromFile(filename + i));
*/
/*
* //Shows game with agents (press Enter to see next game tick)
*
* PlayGame(agents, true);
*/
/*
* //Teach agents by playing yourself versus them.
*
* var agents1 = Teach(agents, 1, 10);
*/
/*
* //Agents learn playing against each other
*
* var agents1 = Gen(agents, 100);
*/
/*
* //Save agents
* var filename = "agent";
* iter = 0;
* foreach (var agent in agents)
* agent.Save(filename + iter++);
*/
///////////////////////////////////////////////////////////////////////////////////////////////////////////
//Here starts actual code (loads pretty fine agents tho)
/*var filename = "agent";
* var iter = GameRules.numberOfAgents;
* var agents = new List<NeuralEnvironment>();
* for(var i = 0; i < iter; i++)
* agents.Add(NetworkCreator.ReadFromFile(filename + i));
*/
//loaded agents
var filename = "agent";
var iter = GameRules.numberOfAgents;
var agents = new List <NeuralEnvironment>();
for (var i = 0; i < iter; i++)
{
agents.Add(NetworkCreator.ReadFromFile(filename + i));
}
//let'em learn for 10 generations and watch their averages
agents = Gen(agents, 10);
//trying to teach'em
agents = Teach(agents, 3, 5);
//10 more generation to see the affect of us being a poor teacher
agents = Gen(agents, 10);
//probly don't wanna save that
/*iter = 0;
* foreach (var agent in agents)
* agent.Save(filename + iter++);
*/
Console.WriteLine("Finished");
Console.ReadLine();
}
