public Neuron(NeuronParam neuronParam)
{
this._metadata = new ExpandoObject();
this.Bias = neuronParam.Bias;
this.Dendrites = new Dendrite[neuronParam.DendriteCount];
for (int d = 0; d < neuronParam.DendriteCount; d++)
{
this.Dendrites[d] = new Dendrite(neuronParam.Weights[d]);
}
}
public NeuralNetwork(int[] layers, double[] weights = null)
{
// set to true if needs to load data
var hasWeight = weights != null;
var readIndex = 0;
// At least we need input and output layers
if (layers.Length < 2)
{
throw new ArgumentException(nameof(layers));
}
// if we need to load weights check the number of them are correct.
if (hasWeight)
{
if (weights.Length != CalculateWeightsCount(layers))
{
throw new ArgumentException(nameof(weights));
}
}
this.Layers = new Layer[layers.Length];
for (int l = 0; l < layers.Length; l++)
{
Layer layer = new Layer(layers[l]);
this.Layers[l] = layer;
for (int n = 0; n < layers[l]; n++)
{
var nDandrites = l > 0 ? layers[l - 1] : 0;
var param = new NeuronParam(nDandrites);
if (l == 0)
{
param.Bias = 0;
}
else
{
if (hasWeight)
{
// Load data from weights
param.Bias = weights[readIndex++];
for (int d = 0; d < nDandrites; d++)
{
param.Weights[d] = weights[readIndex++];
}
}
else
{
// Set random data
param.Bias = new CryptoRandom().RandomValue;
for (int d = 0; d < nDandrites; d++)
{
param.Weights[d] = new CryptoRandom().RandomValue;
}
}
}
layer.Neurons[n] = new Neuron(param);
}
}
}
