/// <summary>
/// </summary>
/// <param name="nm">Режима работы сети (Train, Test, Demo).</param>
/// <param name="LayerInput">Входной слой.</param>
public Network(NetworkMode nm, LayerInput LayerInput)
{
int[] numbNeuronLayers = { 1800, 120, 84, 2 };
input_layer = LayerInput;
Console.WriteLine($"Слой свертки...");
first_convolution = new ConvolutionFirst();
Console.WriteLine($"Слой субдескритизации...");
first_subsamping = new SubsampingFirst();
Console.WriteLine($"Слой свертки...");
second_convolution = new ConvolutionSecond();
Console.WriteLine($"Слой субдескритизации...");
second_subsamping = new SubsampingSecond();
int numbLayerHiddens = 2;
hidden_layer = new LayerHidden[numbLayerHiddens];
for (int i = 0; i < numbLayerHiddens; i++)
{
Console.WriteLine($"Полносвязный слой {i}...");
hidden_layer[i] = new LayerHidden(numbNeuronLayers[i + 1], numbNeuronLayers[i],
NeuronType.Hidden, $"LayerHiddenConvolution{i}");
}
Console.WriteLine($"Полносвязный слой output...");
output_layer = new LayerOutput(numbNeuronLayers[numbNeuronLayers.Length - 1], numbNeuronLayers[numbNeuronLayers.Length - 2],
NeuronType.Output, "LayerOutputConvolution");
fact = new double[numbNeuronLayers[numbNeuronLayers.Length - 1]];
}
public void TestDeserialize()
{
var path = "D:\\1.xml";
var layInput = (LayerInput)LayerInput.Deserialize(path);
Assert.AreEqual(layInput.ShapeIn.Rows, 4);
}
public void TestActivated()
{
var inputLayer = new LayerInput(new DataShape(4, 1));
var(item1, item2) = inputLayer.Activated(new DenseMatrix(4, 1));
Assert.AreEqual(item1, item2);
}
public void TestSerializer()
{
var path = "D:\\1.xml";
var inputLayer = new LayerInput(new DataShape(4, 1));
inputLayer.Serializer(path);
Assert.IsTrue(File.Exists(path));
}
/// <summary>
/// Creates instance of layer input for the specified layer.
/// </summary>
private static IDatasourceInput ReopenLayerInput(ILayer layer, IDatasourceInput oldInput)
{
if (layer == null)
{
return(null);
}
var input = new LayerInput(layer);
var vector = oldInput as IVectorInput;
if (vector != null)
{
input.SelectedOnly = vector.SelectedOnly;
}
return(input);
}
private void AddLayerInput(int neurons)
{
var layer = new LayerInput
{
Neurons = new List <Neuron>(),
};
for (var i = 1; i <= neurons; i++)
{
var neuron = new NeuronInput
{
Connections1 = new List <Connection>(),
Connections2 = new List <Connection>(),
};
neuron.Bias = Math2.Range(-1f, 1);
layer.Neurons.Add(neuron);
}
this.Layers.Add(layer);
}
/// <summary>
/// Конструктор.
/// </summary>
/// <param name="non">Число нейронов текущего слоя.</param>
/// <param name="nopn">Число нейронов предыдущего слоя.</param>
/// <param name="nt">Тип нейрона.</param>
/// <param name="type">Название слоя.</param>
protected Layer(int non, int nopn, NeuronType nt, string type)
{
numofneurons = non;
numofprevneurons = nopn;
Neurons = new Neuron[non];
Console.WriteLine($"Чтение весов");
double[,] Weights = WeightInitialize(MemoryMode.GET, type);
Console.WriteLine($"Чтение весов окончено");
lastdeltaweights = Weights;
for (int i = 0; i < non; ++i)
{
double[] temp_weights = new double[nopn + 1];
for (int j = 0; j < nopn + 1; ++j)
{
temp_weights[j] = Weights[i, j];
}
Neurons[i] = new Neuron(null, temp_weights, nt);
}
output = new LayerInput();
}
public void CreateLayNet(TrainParameters layNetParameter)
{
LayerNet = new List <Layer>();
var inputLayer = new LayerInput(InputDataShape)
{
LayerIndex = 0
};
LayerNet.Add(inputLayer);
layNetParameter.InterLayerStructs.ToList().ForEach(a =>
{
var dataShape = LayerNet.Last().ShapeOut;
switch (a.LayerType)
{
case LayerType.FullConnectLayer:
var fullConnectedLayer = new LayerFullConnected(dataShape, a.NeureCount)
{
LayerIndex = LayerNet.Count
};
LayerNet.Add(fullConnectedLayer);
break;
case LayerType.PoolingLayer:
var poolingLayer = new LayerPooling(dataShape)
{
LayerIndex = LayerNet.Count
};
LayerNet.Add(poolingLayer);
break;
}
});
var outputLayer = new LayerOutput(LayerNet.Last().ShapeOut, OutputDataShape)
{
LayerIndex = LayerNet.Count
};
LayerNet.Add(outputLayer);
}
public void TestInitial()
{
var inputLayer = new LayerInput(new DataShape(4, 1));
Assert.AreNotEqual(inputLayer, null);
}
