public static bool Test_Weights()
{
var cut = MLPNetwork.CreateMLPNetwork(new int[] { 1, 1 }, new double[] { 1.0 }, null);
var result = cut.Calculate(new double[] { 0.5 });
return(Math.Abs(result[0] - 0.5) < 0.001);
}
public static bool Test_MoreWeights_LinearActivation()
{
var cut = MLPNetwork.CreateMLPNetwork(new int[] { 2, 2, 1 }, new double[] { 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 }, MLPNetwork.Identity);
var result = cut.Calculate(new double[] { 0.5, 0.5 });
return(Math.Abs(result[0] - 2) < 0.001);
}
public void VerifyNetwork()
{
var cut = MLPNetwork.CreateMLPNetwork(new int[] { 1, 1 }, new double[] { 1.0 }, null);
var result = cut.Calculate(new double[] { 0.35 });
Assert.Equal(0.35, result[0]);
}
public static bool Test_ReuseNetwork_Sameresult()
{
var cut = MLPNetwork.CreateMLPNetwork(new int[] { 2, 2, 1 }, new double[] { 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 }, MLPNetwork.Identity);
var result = cut.Calculate(new double[] { 0.5, 0.5 });
var result2 = cut.Calculate(new double[] { 0.5, 0.5 });
return(Math.Abs(result[0] - result2[0]) < 0.001);
}
public static bool Test_MoreWeights()
{
var cut = MLPNetwork.CreateMLPNetwork(new int[] { 2, 2, 1 }, new double[] { 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 }, null);
var result = cut.Calculate(new double[] { 0.5, 0.5 });
var expected = Math.Tanh(1.0) + Math.Tanh(1.0);
return(Math.Abs(result[0] - expected) < 0.001);
}
public static bool Test_Unwrap_Create_SameResult()
{
var cut = MLPNetwork.CreateMLPNetwork(new int[] { 1, 1 }, new double[] { 1.0 }, null);
var result1 = cut.Calculate(new double[] { 0.5 });
var weights = cut.ExtractWeights();
var cut2 = MLPNetwork.CreateMLPNetwork(new int[] { 1, 1 }, weights, null);
var result2 = cut2.Calculate(new double[] { 0.5 });
return(Math.Abs(result1[0] - result2[0]) < 0.001);
}
public static MLPNetwork CreateMLPNetwork(NetworkSetup setup)
{
var network = new MLPNetwork {
Layers = new Node[setup.Layers.Count][]
};
for (var i = 0; i < setup.Layers.Count; i++)
{
var layer = setup.Layers[i];
var includeBias = i != setup.Layers.Count - 1;
network.Layers[i] = Enumerable.Range(0, layer.NodeCount + (includeBias?1:0)).Select(l => new Node(layer.ActivationFunction, l == layer.NodeCount)).ToArray();
}
for (var i = 1; i < network.Layers.Length; i++)
{
var weightCounter = 0;
var currentLayer = network.Layers[i];
var previousLayer = network.Layers[i - 1];
var layerSetup = setup.Layers[i];
foreach (var n in previousLayer)
{
foreach (var nextNode in currentLayer)
{
if (nextNode.IsBias)
{
continue;
}
nextNode.InputLinks.Add(new Link(layerSetup.Weights[weightCounter++])
{
SourceNode = n, DestinationNode = nextNode
});
}
}
}
return(network);
}
public NetworkPlayer()
{
var setup = ReadSetup("weights.txt", "network.txt");
m_network = MLPNetworkFactory.CreateMLPNetwork(setup);
}
public NetworkWrapper(string filename, string configName)
{
var setup = ReadSetup(filename, configName);
m_network = MLPNetworkFactory.CreateMLPNetwork(setup);
}
