public void Compute_3Layers()
{
var net = NeuralNetBuilder.From(new NeuralNetConfig()
{
HiddenLayerCount = 3,
HiddenLayerNeuronCount = 2,
InputLabels = new string[] { "A" },
OutputLabels = new string[] { "1" }
});
net.InitializeWeights(new double[2 + 2 + 4 + 4] {
1, 1, // input weights
.5, 1, // neruon 1 to 3 and 4
1, 1, // neruon 2 to 3 and 4
1, 1,
1, 1,
1, 1 // output weights
});
var output = net.Compute(new NamedValue[]
{
new NamedValue()
{
Name = "A", Value = 1
}
});
}
public void CorrectWeightCount()
{
var net = NeuralNetBuilder.From(new NeuralNetConfig()
{
HiddenLayerCount = 1,
HiddenLayerNeuronCount = 1,
InputLabels = new string[] { "A" },
OutputLabels = new string[] { "1" }
});
net.InitializeWeights(new double[2]);
net = NeuralNetBuilder.From(new NeuralNetConfig()
{
HiddenLayerCount = 1,
HiddenLayerNeuronCount = 2,
InputLabels = new string[] { "A" },
OutputLabels = new string[] { "1" }
});
net.InitializeWeights(new double[2 + 2]);
net = NeuralNetBuilder.From(new NeuralNetConfig()
{
HiddenLayerCount = 2,
HiddenLayerNeuronCount = 2,
InputLabels = new string[] { "A" },
OutputLabels = new string[] { "1" }
});
net.InitializeWeights(new double[2 + 2 + 4]);
net = NeuralNetBuilder.From(new NeuralNetConfig()
{
HiddenLayerCount = 3,
HiddenLayerNeuronCount = 2,
InputLabels = new string[] { "A" },
OutputLabels = new string[] { "1" }
});
net.InitializeWeights(new double[2 + 2 + 4 + 4]);
}
public void Compute()
{
var net = NeuralNetBuilder.From(new NeuralNetConfig()
{
HiddenLayerCount = 1,
HiddenLayerNeuronCount = 1,
InputLabels = new string[] { "A" },
OutputLabels = new string[] { "1", "2", "3" }
});
net.InitializeWeights(new double[1 + 3] {
.5, .5, .25, 1
});
var output = net.Compute(new NamedValue[]
{
new NamedValue()
{
Name = "A", Value = 1
}
});
}
public void Sin()
{
var rand = new Random();
var net = NeuralNetBuilder.From(new NeuralNetConfig()
{
InputLabels = new string[] { "x" },
OutputLabels = new string[] { "y" },
HiddenLayerCount = 3,
HiddenLayerNeuronCount = 3,
//PreActivationFunction = (i, d) =>
ActivationFunction = d => d > .5 ? 1 : 0
//ActivationFunction = d => 1 / (1 + Math.Pow(Math.E, -d))
});
var trainingData = new List <Tuple <double, double> >();
for (double i = 0; i < Math.PI * 2; i += .1)
{
var input = i;
var output = Math.Sin(i);
//var normalizedOutput = output / Math.PI * 2;
var normalizedOutput = .5 * (output + 1);
var normalizedInput = input / (Math.PI * 2);
trainingData.Add(new Tuple <double, double>(normalizedInput, normalizedOutput));
}
var weightCount = net.WeightCount;
var biasCount = net.BiasCount;
var weightOrgSpawner = new SpawnFunction <double[]>(index =>
{
var data = new double[weightCount + biasCount];
for (int i = 0; i < weightCount + biasCount; i++)
{
var scale = i >= weightCount ? 1 : 1;
data[i] = (rand.NextDouble()) * scale;
}
return(new Org(data));
});
var fitnessFunction = new FitnessFunction <double[]>(org =>
{
net.InitializeWeights(org.Genome.Take(weightCount).ToArray());
net.InitializeBiases(org.Genome.Skip(weightCount).ToArray());
var errorSum = trainingData.Select(trainingElement =>
{
var input = trainingElement.Item1;
var expectedOutput = trainingElement.Item2;
var actualOutput = net.Compute(new NamedValue[] { new NamedValue()
{
Name = "x", Value = input
} });
var error = expectedOutput - actualOutput[0].Value;
var convertedError = Math.Pow((Math.Abs(error) + 1), 2);
return(convertedError);
}).Sum();
//Trace.WriteLine(errorSum);
return(1 / errorSum);
});
var geneticSim = new GeneticSimulation <double[]>(fitnessFunction, weightOrgSpawner, 50);
var bestSet = geneticSim.Evolve(20000, false);
var theBest = geneticSim.GetBest();
net.InitializeWeights(theBest.Genome.Take(weightCount).ToArray());
net.InitializeBiases(theBest.Genome.Skip(weightCount).ToArray());
Trace.WriteLine("OUTPUT");
trainingData.ForEach(t =>
{
var input = t.Item1;
var expectedOutput = t.Item2;
var actualOutput = net.Compute(new NamedValue[] { new NamedValue()
{
Name = "x", Value = input
} });
var error = expectedOutput - actualOutput[0].Value;
//Trace.WriteLine($"INPUT:{input} EXPECT:{expectedOutput} REAL:{actualOutput[0].Value} ERR:{error}");
//Trace.WriteLine($"{input},{expectedOutput},{actualOutput[0].Value}");
Trace.WriteLine($"{actualOutput[0].Value}");
});
}
