public void Run()
{
Network = new NeuralNetwork(NetworkProperties.Default, 2, 3, 1);
Network.Initialize();
Console.WriteLine("Enter number of training iterations:");
int trainingIterations = int.Parse(Console.ReadLine());
TrainingPair[] trainingSet = new TrainingPair[]
{
new TrainingPair(new Vector(0, 0), new Vector(0f)),
new TrainingPair(new Vector(0, 1), new Vector(1f)),
new TrainingPair(new Vector(1, 0), new Vector(1f)),
new TrainingPair(new Vector(1, 1), new Vector(0f))
};
Stopwatch time = Stopwatch.StartNew();
for (int i = 0; i < trainingIterations; i++)
{
Random r = new Random();
TrainingPair set = trainingSet[r.Next(0, trainingSet.Length)];
Network.Train(set.Input, set.Output);
Console.WriteLine();
Network.Input.Nodes.Print("Input");
Network.Output.Nodes.Print("Output");
}
long dt = time.ElapsedMilliseconds;
Console.WriteLine("Training Time: " + dt + "ms");
Console.WriteLine("Training Complete\n");
while (true)
{
Vector input = new Vector(2);
Console.WriteLine("Enter Input 1:");
float input1 = float.Parse(Console.ReadLine());
Console.WriteLine("Enter Input 2:");
float input2 = float.Parse(Console.ReadLine());
input[0] = input1;
input[1] = input2;
input.Print("Input Vector");
Network.Input.Nodes = input;
Network.FeedForward();
Network.Output.Nodes.Print("Results");
Console.WriteLine("Exit? (y/n)");
if (Console.ReadLine() == "y")
{
break;
}
}
}
/// <summary>
/// Writes
/// </summary>
/// <param name="debug">If set to <c>true</c>, write all recorded </param>
public void WriteToFile(bool debug = false)
{
TrainingPair trainingPair = new TrainingPair(this.goal);
trainingPair.InitializeFromGame(this.self, this.self.transform.FindChild("FirstPersonCharacter").gameObject);
// write to file with features and target movements
if (debug)
{
//debug.Log (trainingPair.ToString ());
}
file.WriteLine(trainingPair.ToString());
file.Flush();
}
void Start()
{
var patterns = new TrainingPair[] {
new TrainingPair()
{
inputs = new double[] { -1, -1 }, result = new double[] { -1 }
},
new TrainingPair()
{
inputs = new double[] { -1, +1 }, result = new double[] { +1 }
},
new TrainingPair()
{
inputs = new double[] { +1, -1 }, result = new double[] { +1 }
},
new TrainingPair()
{
inputs = new double[] { +1, +1 }, result = new double[] { -1 }
}
};
Train(patterns, 10000);
Test(patterns);
}
void Train(TrainingPair[] patterns, int iter=100, double N=0.5, double M=0.1)
{
for (var i=0; i<iter; i++) {
var error = 0.0;
foreach (var p in patterns) {
var inputs = p.inputs;
var targets = p.result;
Calculate (inputs);
error = error + BackPropagate (targets, N, M);
}
}
}
void Test(TrainingPair[] patterns)
{
foreach (var p in patterns) {
Debug.Log (p);
var log = "Actual:" + string.Join(" ", (from i in Calculate (p.inputs) select i.ToString()).ToArray());
Debug.Log (log);
}
}
void Start()
{
var patterns = new TrainingPair[] {
new TrainingPair () { inputs=new double[] {-1,-1}, result=new double[] { -1 }},
new TrainingPair () { inputs=new double[] {-1,+1}, result=new double[] { +1 }},
new TrainingPair () { inputs=new double[] {+1,-1}, result=new double[] { +1 }},
new TrainingPair () { inputs=new double[] {+1,+1}, result=new double[] { -1 }}
};
Train (patterns, 10000);
Test (patterns);
}
public void TrainAINetwork(string inputPath, string serializePath = null)
{
if (serializePath != null && File.Exists(serializePath))
{
using (FileStream fs = new FileStream(serializePath, FileMode.Open)) {
this.network = (ActivationNetwork) new BinaryFormatter().Deserialize(fs);
return;
}
}
List <TrainingPair> trainingList = new List <TrainingPair>();
System.IO.StreamReader file =
new System.IO.StreamReader(inputPath);
string line;
List <List <double> > input = new List <List <double> >();
List <List <double> > output = new List <List <double> > ();
// skip the header (first line);
file.ReadLine();
while ((line = file.ReadLine()) != null)
{
TrainingPair tp = new TrainingPair(this.goal);
line = line.Trim();
tp.InitializeFromSaved(line);
trainingList.Add(tp);
if (tp.observedAction.xRotInput != 0.0f || tp.observedAction.yRotInput != 0.0f || tp.observedAction.forwardPan != 0.0f ||
tp.observedAction.horizontalPan != 0.0f || tp.observedAction.fireButtonDown != 0.0f)
{
// only add non-zero examples for now
input.Add(new List <double> ()
{
tp.gameStateSummary.XZAngleToObj,
tp.gameStateSummary.YZAngletoObj,
tp.gameStateSummary.distToObj
});
output.Add(new List <double> ()
{
tp.observedAction.yRotInput,
tp.observedAction.xRotInput,
tp.observedAction.horizontalPan,
tp.observedAction.forwardPan,
tp.observedAction.fireButtonDown,
tp.observedAction.sprintButtonDown,
tp.observedAction.jumpButtonDown
});
}
}
file.Close();
Debug.Log(string.Format("Training List Length: {0}", input.Count));
for (int i = 0; i < Config.Instance.node["training_epochs"].AsInt; i++)
{
double error = this.teacher.RunEpoch(NestedListToArray(input), NestedListToArray(output));
if (i % 50 == 0)
{
Debug.Log(string.Format("iteration: {0}, Error: {1}", i, error));
}
}
if (serializePath != null)
{
using (FileStream fs = new FileStream(serializePath, FileMode.Create))
{
new BinaryFormatter().Serialize(fs, this.network);
}
}
}