public void CreateStandardMLANetwork(int iil, int oil, int nc, int nl)
{
PexAssume.InRange(iil, 1, 41);
PexAssume.InRange(oil, 1, 41);
PexAssume.InRange(nc, 1, 41);
PexAssume.InRange(nl, 2, 41);
var nodeF = new Factory<NeuralNode>(() => new ActivationNeuron(new SigmoidActivationFunction(1.05)));
var connF = new Factory<NeuralConnection>(() => new Synapse());
var layers = new NeuralLayerDefinition[nl];
for (int i = 0; i < nl; i++)
{
if (i < (nl - 1))
{
layers[i] = new NeuralLayerDefinition(nodeF, nc);
}
else
{
layers[i] = new NeuralLayerDefinition(nodeF, oil);
}
}
var a = new StandardMultilayerArchitecture(new NeuralConnectionDefinition(connF, false), iil, layers);
var net = a.CreateNetwork();
net.Iteration();
}
private static void TestNN()
{
var arch = new StandardMultilayerArchitecture
{
InitParameters = new OpenCLNNInitParameters
{
},
InputSize = 10,
HiddenLayers = new[]
{
new ActivationLayer(10, new SigmoidActivationFunction())
},
OutputLayer = new ActivationLayer(2, new LinearActivationFunction())
};
using (var nn = (OpenCLNeuralNetwork)arch.CreateNetwork())
{
float[] inputs = new float[10];
float[] outputs = new float[2];
nn.WriteInput(inputs);
nn.ReadOutput(outputs);
}
}
public void CanCreateSMLCPUTest()
{
var arch = new StandardMultilayerArchitecture
{
InitParameters = new CPUNNInitParameters(),
InputSize = 10,
HiddenLayers = new[]
{
new ActivationLayer(10, new SigmoidActivationFunction())
},
OutputLayer = new ActivationLayer(2, new LinearActivationFunction())
};
var nn = arch.CreateNetwork();
var cpuNN = nn as CPUNeuralNetwork;
Assert.IsNotNull(nn);
Assert.IsNotNull(cpuNN);
}
private static NeuralNetworkArchitecture CreateArch(NNInitParameters initPars)
{
var arch = new StandardMultilayerArchitecture
{
InitParameters = initPars,
InputSize = 10,
HiddenLayers = new[]
{
new ActivationLayer(200, new SigmoidActivationFunction(), new GradientDescentRule()),
new ActivationLayer(100, new SigmoidActivationFunction(), new GradientDescentRule())
},
OutputLayer = new ActivationLayer(20, new LinearActivationFunction(), new GradientDescentRule())
};
return arch;
}
