public void XOrTest()
{
int times = 100000;
NeuralNetworkTrainer trainer = new NeuralNetworkTrainer();
trainer.HiddenLayers = new int[] { 2 };
trainer.lambda = 0;
trainer.AddSample(new BinaryClassificationSample(new double[] { 0, 0 }, 0))
.AddSample(new BinaryClassificationSample(new double[] { 0, 1 }, 1))
.AddSample(new BinaryClassificationSample(new double[] { 1, 0 }, 1))
.AddSample(new BinaryClassificationSample(new double[] { 1, 1 }, 1));
trainer.Train(times);
for (int i = 0; i < 9; i++)
{
Debug.Print(trainer.costs[i].ToString());
}
for (int i = 0; i < times; i += times / 100)
{
Debug.Print(trainer.costs[i].ToString());
}
NeuralNetworkClassifier classifier = new NeuralNetworkClassifier(trainer.Theta);
Assert.AreEqual(0, classifier.Predict(0, 0));
Assert.AreEqual(1, classifier.Predict(0, 1));
Assert.AreEqual(1, classifier.Predict(1, 1));
Assert.AreEqual(1, classifier.Predict(1, 1));
}
void Start()
{
startLine = GameObject.FindGameObjectWithTag("Start line").transform;
networkTrainer = new NeuralNetworkTrainer(folder, template, numberOfAiClients);
CreateAiClients();
ResetAiClients();
}
public void CostTest()
{
Matrix theta1 = new double[, ] {
{ 0.5, 1, 1 }, { 1, 1, 1 }
};
Matrix theta2 = new double[, ] {
{ 1, 1, 0.1 }
};
Matrix[] theta = new Matrix[] { theta1, theta2 };
Matrix x = new double[, ] {
{ 0, 0 }, { 0, 1 }, { 1, 0 }, { 1, 1 }
};
Vector y = new double[] { 0, 1, 1, 0 };
var costf = NeuralNetworkTrainer.CostFunction(theta, x, y, 0);
double cost = costf.Item1;
var t = costf.Item2;
Debug.Print(cost.ToString());
foreach (var tt in t)
{
Debug.Print(tt.ToString());
}
}
public static void Main(string[] args)
{
//read example text
string exampleTxt = File.ReadAllText("example.txt");
File.Create("output.txt");
//encode in one-hot encoding, each letter is assigned a unique neuron.
List <char> dict = null;
float[][] data = Utils.EncodeStringOneHot(exampleTxt, out dict);
//create neural network
int hlen = exampleTxt.Length / 4; // dict.Count * 2;
NeuralNetwork neuralNetwork = new NeuralNetwork(new NeuralNetworkLayer(dict.Count, false, null), //input layer
new NeuralNetworkLayer[] { //hidden layers
new NeuralNetworkLayer(hlen, true, Utils.Rectifier_ActivationFunction),
},
new NeuralNetworkLayer(dict.Count, false, Utils.Rectifier_ActivationFunction)//output layer
);
neuralNetwork.RandomizeWeightsAndBiases(0.0f, 0.0f, 0.0f, 0.5f / hlen);
//convert into training data for neural network
//we want to input the current text character and have the neural network output the next text character.
float[][] inputDat = new float[data.Length - 1][],
targetDat = new float[data.Length - 1][];
for (int i = 0; i < data.Length - 1; i++)
{
inputDat[i] = data[i];
targetDat[i] = data[i + 1];
}
//train neural network and every 10 seconds output text generated from the network
NeuralNetworkTrainer trainer = new NeuralNetworkTrainer(neuralNetwork, inputDat, targetDat, 2, NeuralNetworkTrainer.LOSS_TYPE_AVERAGE);
trainer.learningRate = 0.01f;
trainer.desiredLoss = 0.0f;
trainer.lossSmoothing = 0.0f;
//program for generating text
NeuralNetworkProgram program = new NeuralNetworkProgram(neuralNetwork);
//init trainer for training
trainer.StartInit();
long lastTime = 0,
nowTime;
while (true)
{
trainer.Learn();
nowTime = System.DateTime.Now.Ticks / 10000;
if (nowTime - lastTime > 10000)
{
//generate text(100-200 chars) from network
string txt = "";
program.context.Reset(true);
float[] id = program.context.inputData,
od = program.context.outputData;
int currentChar = Utils.NextInt(0, dict.Count), //random starting character
nChars = Utils.NextInt(100, 200); //between 100-200 characters
txt += dict[currentChar];
for (int i = 0; i < nChars; i++)
{
//input current character to network
id[currentChar] = 1.0f;
//run network
program.Execute();
//manually clear last character from input
id[currentChar] = 0.0f;
//treat output characters as probabilities, randomly sampling next character from probabilities
Utils.Normalize(od);
currentChar = Utils.RandomChoice(od);
txt += dict[currentChar];
}
//output iterations, loss and generated text
System.Console.WriteLine("Iterations: " + trainer.GetIterations() + ", Loss: " + trainer.GetLoss() + System.Environment.NewLine +
"Text: " + txt + System.Environment.NewLine);
string outputOld = File.ReadAllText("output.txt");
string output = outputOld + System.Environment.NewLine + "Iterations: " + trainer.GetIterations() + ", Loss: " + trainer.GetLoss() + System.Environment.NewLine +
"Text: " + txt + System.Environment.NewLine;
File.WriteAllText("output.txt", output);
lastTime = nowTime;
}
}
}