private void InitializeNN(List <Tuple <int, NeuronType> > layout)
{
_nn = new FFNeuralNetwork(new Random());
_nn.Initialize(layout);
NNVis.Initialize(_nn);
NN_Export_btn.IsEnabled = true;
NN_toolBar.IsEnabled = true;
}
public void Update(FFNeuralNetwork nn)
{
if (NNVisSettings.EnableNNVis)
{
UpdateNNVis(nn);
}
if (NNVisSettings.EnableNNErrorVis)
{
UpdateNNErrorVis(nn);
}
}
private void InitializeNN(JObject jnn)
{
_nn = new FFNeuralNetwork(new Random());
if (!_nn.Initialize(jnn))
{
_nn = null;
return;
}
NNVis.Initialize(_nn);
NN_Export_btn.IsEnabled = true;
NN_toolBar.IsEnabled = true;
}
private void InitializeNNVis(FFNeuralNetwork nn)
{
Canvas.Children.Clear();
_nodes = new List <List <NNVis_Node> >();
var inputs = new List <NNVis_Node>();
for (var i = 0; i < nn.InputLayerSize; i++)
{
inputs.Add(new NNVis_Input(Canvas, new Point(10 + NNVisSettings.NodeHSpace, 10 + 50 + i * NNVisSettings.NodeVSpace)));
}
_nodes.Add(inputs);
_weights = new List <List <List <NNVis_Weight> > >();
for (var l = 0; l < nn.Network.Count; l++)
{
var layer = new List <NNVis_Node>();
var weightstmp = new List <List <NNVis_Weight> >();
for (var n = 0; n < nn.Network[l].Count; n++)
{
var node = new NNVis_Neuron(Canvas,
new Point((_nodes.Count + 1) * NNVisSettings.NodeHSpace, 50 + n * NNVisSettings.NodeVSpace));
node.SetOutput(nn.Network[l][n].Output);
node.SetBias(nn.Network[l][n].Bias);
node.SetError(nn.Network[l][n].Error);
layer.Add(node);
var weightstmp2 = new List <NNVis_Weight>();
for (var w = 0; w < _nodes[l].Count; w++)
{
var weight = new NNVis_Weight(Canvas, _nodes[l][w].GetWeightPosition(true),
layer[n].GetWeightPosition(false));
weight.SetValue(nn.Network[l][n].Weights[w]);
weightstmp2.Add(weight);
}
weightstmp.Add(weightstmp2);
}
_nodes.Add(layer);
_weights.Add(weightstmp);
}
RescaleCanvas();
UpdateSettings();
}
private void UpdateNNVis(FFNeuralNetwork nn)
{
for (var l = 0; l < nn.Network.Count; l++)
{
for (var n = 0; n < nn.Network[l].Count; n++)
{
_nodes[l + 1][n].SetOutput(nn.Network[l][n].Output);
_nodes[l + 1][n].SetError(nn.Network[l][n].Error);
_nodes[l + 1][n].SetBias(nn.Network[l][n].Bias);
for (var w = 0; w < nn.Network[l][n].Weights.Count; w++)
{
_weights[l][n][w].SetValue(nn.Network[l][n].Weights[w]);
}
}
}
}
private void UpdateNNErrorVis(FFNeuralNetwork nn)
{
_error.Add(new Point(_error.Count, nn.Error));
if (_error.Count > NNVisSettings.MaxErrorPoints)
{
_error.RemoveAt(0);
var pc = new PointCollection();
for (var x = 0; x < _error.Count; x++)
{
pc.Add(new Point(x, _error[x].Y));
}
_error = pc;
}
Plot.Add("Error", _error, Colors.White);
IterationLabel.Content = $"I: {nn.Iteration}";
ErrorLabel.Content = $"E: {nn.Error:0.#######}";
}
public void Initialize(FFNeuralNetwork nn)
{
InitializeNNVis(nn);
InitializeNNErrorVis();
}