public static void Save(string fileName, NeuralNet neuralNet, int epoch, double mse, double initialMse, double accuracy)
{
var nns = new NeuralNetSave()
{
Rate = neuralNet.Parameters.Rate,
Momentum = neuralNet.Parameters.Momentum,
Activation = neuralNet.Parameters.Activation,
ValidationAccuracy = accuracy,
SnapshotInterval = neuralNet.Parameters.SnapshotInterval,
BatchSize = neuralNet.Parameters.BatchSize,
Mse = mse,
InitialMse = initialMse,
Epoch = epoch,
Layers = new List <Layer>()
};
foreach (var layer in neuralNet.Layers)
{
var newLayer = new Layer()
{
Type = layer.Type,
Nodes = new List <Node>()
};
foreach (var node in layer.Nodes)
{
var newNode = new Node()
{
Weights = node.Weights
};
switch (layer.Type)
{
case NeuralNet.LayerType.Input:
newNode.Feature = ((NeuralNet.InputNode)node).Feature;
break;
case NeuralNet.LayerType.Output:
newNode.IsContinuous = ((NeuralNet.OutputNode)node).IsContinuous;
newNode.LabelCol = ((NeuralNet.OutputNode)node).LabelCol;
newNode.LabelVal = ((NeuralNet.OutputNode)node).LabelVal;
break;
}
newLayer.Nodes.Add(newNode);
}
nns.Layers.Add(newLayer);
}
var json = JsonConvert.SerializeObject(nns, Formatting.Indented);
using (var w = new StreamWriter(fileName))
{
w.Write(json);
}
}
public static void Load(string filePath, NeuralNet neuralNet)
{
var r = new StreamReader(filePath);
var json = r.ReadToEnd();
var ms = new MemoryStream(Encoding.UTF8.GetBytes(json));
var ser = new DataContractJsonSerializer(typeof(NeuralNetSave));
var nns = ser.ReadObject(ms) as NeuralNetSave;
ms.Close();
neuralNet.Parameters.Rate = nns.Rate;
neuralNet.Parameters.Momentum = nns.Momentum;
neuralNet.Parameters.Activation = nns.Activation;
neuralNet.Parameters.SnapshotInterval = nns.SnapshotInterval;
neuralNet.Parameters.BatchSize = nns.BatchSize;
neuralNet.Parameters.StartEpoch = nns.Epoch;
neuralNet.Parameters.InitialMse = nns.InitialMse;
neuralNet.Parameters.StartMse = nns.Mse;
neuralNet.Layers = new List <NeuralNet.Layer>();
NeuralNet.Layer prevLayer = null;
foreach (var layer in nns.Layers)
{
var newLayer = new NeuralNet.Layer()
{
Type = layer.Type,
Nodes = new List <NeuralNet.Node>()
};
var idx = 0;
foreach (var node in layer.Nodes)
{
NeuralNet.Node newNode = null;
switch (layer.Type)
{
case NeuralNet.LayerType.Input:
newNode = new NeuralNet.InputNode(idx, node.Feature, null);
break;
case NeuralNet.LayerType.Hidden:
newNode = new NeuralNet.HiddenNode(idx, node.Weights);
break;
case NeuralNet.LayerType.Output:
newNode = new NeuralNet.OutputNode(idx, node.IsContinuous, node.LabelCol, node.LabelVal, node.Weights);
break;
}
idx++;
newLayer.Nodes.Add(newNode);
}
newLayer.Previous = prevLayer;
neuralNet.Layers.Add(newLayer);
if (prevLayer != null)
{
prevLayer.Next = newLayer;
}
prevLayer = newLayer;
}
}