public virtual INeuralNetwork CreateNeuralNetwork(int inputNodes, int hiddenNodes, int staticWeights=0)
{
var wrapper = new AForgeNeuralNetwork(inputNodes, hiddenNodes,staticWeights);
wrapper.Randomize();
return wrapper;
}
/// <summary>
/// Initialise to random values, implemented by generating a new network and extracting the relevant weights
/// </summary>
/// <returns></returns>
private double[] DefaultNeuronWeights()
{
double[] weights = null;
var net =
new AForgeNeuralNetwork(MorphoSyntacticContext.VectorLength + MorphoSyntacticContext.VectorLength, MorphoSyntacticContext.VectorLength);
net.Randomize();
net.Theta.SaveTo(ref weights);
return weights;
}
public void TestStaticWeightsPreserved()
{
double[] weights = null;
_target.SaveTo(ref weights);
var net2 = new AForgeNeuralNetwork(4, 5, 2);
net2.Theta.LoadFrom(weights);
var result1 = _net1.Run(new[] { 0.3, 0.4, 0.0, 0.0});
var result2 = net2.Run(new[] { 0.3, 0.4, 0.0, 0.0 });
_net1.HiddenLayer[0].Should().NotBe(net2.HiddenLayer[0],"Static weights from neurons were not preserved after LoadFrom");
}
public void TestNonStaticWeightsRestored()
{
double[] weights = null;
_target.SaveTo(ref weights);
var net2 = new AForgeNeuralNetwork(4, 5, 2);
net2.Theta.LoadFrom(weights);
_net1.UnderlyingNetwork.Layers[0].Neurons[0].Weights[3].Should().Be(net2.UnderlyingNetwork.Layers[0].Neurons[0].Weights[3]);
var result1 = _net1.Run(new[] { 0.0, 0.0, 0.4, 0.5});
var result2 = net2.Run(new[] { 0.0, 0.0, 0.4, 0.5});
_net1.HiddenLayer[0].Should().Be(net2.HiddenLayer[0], "same inputs with restored weights");
}
public void Setup()
{
_net1 = new AForgeNeuralNetwork(4, 5, 2);
_target = _net1.Theta as AForgeTheta;
}
