public static Network Load(string path, double learnRate)
{
if (File.Exists(path))
{
NetworkSaveModel model = JsonHelper.Deserialize <NetworkSaveModel>(path);
Network net = new Network(model.Input, model.Hidden, model.Output, learnRate);
int lastLayerNumber = model.Hidden.Length + 1; //output layer number - 1
foreach (var weight in model.Weights)
{
string[] parts = weight.Key.Split('_');
//1. in layer number
//2. in neuron number
//3. out neuron number
if (parts.Length != 3)
{
continue;
}
int[] numeric = new int[parts.Length];
for (int i = 0; i < parts.Length; i++)
{
numeric[i] = int.Parse(parts[i]);
}
if (numeric[0] == 0) //input layer neurons
{
net.inputLayer.Neurons[numeric[1]].OutBoundConnection(numeric[2]).weight = weight.Value;
}
else //layer connected to output
{
net.hiddenLayers[numeric[0] - 1].Neurons[numeric[1]].OutBoundConnection(numeric[2]).weight = weight.Value;
}
}
return(net);
}
return(null);
//if (File.Exists(path))
//{
// Network n = JsonHelper.Deserialize<Network>(path);
// return n;
//}
//return null;
}
public void Save(string path)
{
NetworkSaveModel save = new NetworkSaveModel();
save.Input = inputLayer.Size - 1;
save.Output = outputLayer.Size;
save.Hidden = new int[hiddenLayers.Length];
for (int i = 0; i < hiddenLayers.Length; i++)
{
save.Hidden[i] = hiddenLayers[i].Size - 1;
}
List <Layer> layers = new List <Layer>();
layers.Add(inputLayer);
foreach (var hidden in hiddenLayers)
{
layers.Add(hidden);
}
layers.Add(outputLayer);
save.Weights = new Dictionary <string, double>();
for (int i = 0; i < layers.Count - 1; i++)
{
//LAYER_NEURON_LAYER_NEURON
string key_template = "{0}_{1}_{2}";
for (int j = 0; j < layers[i].Neurons.Length; j++)
{
for (int k = 0; k < layers[i].Neurons[j].OutBoundSize; k++)
{
string key = string.Format(key_template, i, j, k);
save.Weights.Add(key, layers[i].Neurons[j].OutBoundConnection(k).weight);
}
}
}
JsonHelper.Serialize(save, path);
//JsonHelper.Serialize(this, path);
}
