/// <summary>
/// Обновить данные слоя.
/// </summary>
/// <param name="updatedFilterCore">Обновлённое ядро фильтра.</param>
/// <param name="updatedInputLayerData">Обновлённый входной слой.</param>
public void UpdateData(double[,] updatedFilterCore,
Dictionary <string, double> updatedInputLayerData)
{
var offset = MatrixConstants.MATRIX_SIZE - MatrixConstants.FILTER_MATRIX_SIZE;
var step = MatrixConstants.MATRIX_SIZE - offset;
var updatedConvolutionalLayerNeurons = new List <NeuronModel>();
var neuronIndex = 0;
for (var xIndex = 0; xIndex < step; ++xIndex)
{
for (var yIndex = 0; yIndex < step; ++yIndex)
{
var inputs = new List <double>();
var weights = new List <double>();
GetDataByFilterCore(updatedFilterCore, xIndex, yIndex, inputs, weights);
var neuron = new NeuronModel(inputs, weights)
{
LastWeights = _convolutionalLayerData[neuronIndex].LastWeights
};
++neuronIndex;
}
}
}
private void RenderNeuronInfo(NeuronModel neuron, Vector3 screenPosition)
{
screenPosition.X += 10;
screenPosition.Y += 10;
using (Blender.AveragingBlender())
{
GL.PushMatrix();
GL.Translate(screenPosition);
GL.Color4(new Color4(100, 100, 100, 220));
GL.Begin(PrimitiveType.Quads);
GL.Vertex3(0, 0, 0);
GL.Vertex3(100, 0, 0);
GL.Vertex3(100, 30, 0);
GL.Vertex3(0, 30, 0);
GL.End();
GL.PopMatrix();
QFont.Begin();
GL.Translate(10, 0, 0);
GL.Translate(screenPosition);
m_font.Print($"{neuron.RegionModel.Index} : {neuron.Index}", QFontAlignment.Left);
QFont.End();
}
}
/// <summary>
/// Инициализация слоёв нейронной сети.
/// </summary>
/// <param name="listOfPicturesmatrix">Список матриц изображений.</param>
/// <param name="layers">Список слоёв.</param>
/// <param name="filterCore">Ядро фильтра.</param>
/// <param name="inputLayerNeurons">Нейроны выходного слоя.</param>
/// <param name="convolutionalLayerNeurons">Нейроны свёрточного слоя.</param>
/// <param name="hiddenLayerNeurons">Нейроны скрытого слоя.</param>
/// <param name="outputNeuron">Нейроны выходного слоя.</param>
private static void LayersInitialize(List <double[, ]> listOfPicturesmatrix, List <Layer> layers,
double[,] filterCore, out Dictionary <string, double> inputLayerNeurons,
out List <NeuronModel> convolutionalLayerNeurons, out List <NeuronModel> hiddenLayerNeurons,
out NeuronModel outputNeuron)
{
var firstDataSet = listOfPicturesmatrix.First();
var inputLayer = new InputLayer(firstDataSet);
inputLayer.Initialize();
layers.Add(inputLayer);
inputLayerNeurons = inputLayer.GetLayerNeurons();
var convolutionalLayer = new ConvolutionalLayer(inputLayerNeurons);
convolutionalLayer.Initialize(filterCore);
layers.Add(convolutionalLayer);
convolutionalLayerNeurons = convolutionalLayer.GetLayerNeurons();
var hiddenLayer = new HiddenLayer(convolutionalLayerNeurons);
hiddenLayer.Initialize();
layers.Add(hiddenLayer);
hiddenLayerNeurons = hiddenLayer.GetLayerNeurons();
var outputLayer = new OutputLayer(hiddenLayerNeurons);
outputLayer.Initilize();
layers.Add(outputLayer);
outputNeuron = outputLayer.GetOutputNeuron();
}
/// <summary>
/// Инициализацяи свёрточного слоя.
/// </summary>
/// <param name="filterCore">Ядро фильтра.</param>
public void Initialize(double[,] filterCore)
{
_convolutionalLayerData = new List <NeuronModel>();
var offset = MatrixConstants.MATRIX_SIZE - MatrixConstants.FILTER_MATRIX_SIZE;
var step = MatrixConstants.MATRIX_SIZE - offset;
var emptyLastWeights = new List <double>();
for (var index = 0; index < MatrixConstants.MATRIX_SIZE *
MatrixConstants.MATRIX_SIZE; ++index)
{
emptyLastWeights.Add(0d);
}
for (var xIndex = 0; xIndex < step; ++xIndex)
{
for (var yIndex = 0; yIndex < step; ++yIndex)
{
var inputs = new List <double>();
var weights = new List <double>();
GetDataByFilterCore(filterCore, xIndex, yIndex, inputs, weights);
var neuron = new NeuronModel(inputs, weights)
{
LastWeights = emptyLastWeights
};
_convolutionalLayerData.Add(neuron);
}
}
}
public void RepositionsSynapse()
{
RegionModel addedRegion;
SynapseModel addedSynapse;
SetupRegionWithSynapse(out addedRegion, out addedSynapse);
NeuronModel neuron1 = m_model.Regions.First().Neurons.First();
NeuronModel neuron2 = m_model.Regions.First().Neurons.Last();
SynapseModel synapse = m_model.Regions.First().Synapses.First();
neuron1.Position *= 2;
ResponseMessage diff = ModelResponseBuilder.Build(repositionedNeurons: new List <NeuronModel> {
neuron1
});
ApplyModelDiff(diff);
Assert.Equal(neuron1.Position, synapse.Position);
neuron2.Position *= 3;
diff = ModelResponseBuilder.Build(repositionedNeurons: new List <NeuronModel> {
neuron2
});
ApplyModelDiff(diff);
Assert.Equal(neuron2.Position, neuron1.Position + synapse.TargetPosition);
}
/// <summary>
/// Обновить входные значения.
/// </summary>
/// <param name="inputs">Входные значения.</param>
public void UpdateData(List <double> inputs)
{
_outputNeuron = new NeuronModel(inputs, _outputNeuron.Weights)
{
LastWeights = _outputNeuron.LastWeights
};
_output = _outputNeuron.Output;
}
public void CorrectlyRepositionsNeurons()
{
var size = new Vector3(10, 20, 30);
var region = new RegionModel(1, "foo", "bar", Vector3.Zero, size);
var neuron1 = new NeuronModel(1, "foo", region, new Vector3(0, 0, 0));
var neuron2 = new NeuronModel(1, "foo", region, new Vector3(1, 1, 1));
Assert.Equal(-size / 2 + new Vector3(RegionModel.RegionMargin), neuron1.Position);
Assert.Equal(size / 2 - new Vector3(RegionModel.RegionMargin), neuron2.Position);
}
private void ToggleNeuron(NeuronModel neuron)
{
if (neuron.Picked)
{
DeselectNeuron(neuron.RegionModel.Index, neuron.Index);
}
else
{
SelectNeuron(neuron.RegionModel.Index, neuron.Index);
}
}
private void SetupRegionWithNeuron(out RegionModel addedRegion, out NeuronModel addedNeuron)
{
addedRegion = SetupRegion(m_applier, m_model, 1);
addedNeuron = new NeuronModel(1, "neuronType", addedRegion, Vector3.One);
ResponseMessage diff = ModelResponseBuilder.Build(addedNeurons: new List <NeuronModel> {
addedNeuron
});
ApplyModelDiff(diff);
}
private bool TryFindNeuron(uint regionIndex, uint neuronIndex, out NeuronModel neuron)
{
RegionModel region;
if (!m_simulationModel.Regions.TryGetModel(regionIndex, out region))
{
neuron = null;
return(false);
}
return(region.Neurons.TryGetModel(neuronIndex, out neuron));
}
/// <summary>
/// Приготовить к распознаванию.
/// </summary>
/// <param name="neuronIndexToWeightsValueDictionary">Словарь значений весов,
/// где ключ - индекс нейрона, значение - веса нейрона.</param>
public void RecognizeMode(Dictionary <int, List <double> > neuronIndexToWeightsValueDictionary)
{
var inputs = new List <double>();
if (!neuronIndexToWeightsValueDictionary.TryGetValue(0, out var weights))
{
ErrorHelper.GetDataError();
}
_hiddenLayerData.ForEach(neuron => inputs.Add(neuron.Output));
_outputNeuron = new NeuronModel(inputs, weights);
}
/// <summary>
/// Обновление весов нейрона.
/// </summary>
/// <param name="neuronModel">Модель нейрона.</param>
/// <param name="weights">Новые веса.</param>
public static void UpdateWeights(this NeuronModel neuronModel, List <double> weights)
{
if (neuronModel.Weights.Count != weights.Count)
{
MessageBox.Show("Не соответствие по количеству весов!", "Ошибка",
MessageBoxButton.OK, MessageBoxImage.Error);
return;
}
neuronModel.LastWeights = neuronModel.Weights;
neuronModel.Weights = weights;
}
/// <summary>
/// Обновление весов нейрона.
/// </summary>
/// <param name="neuronModel">Модель нейрона.</param>
/// <param name="weights">Новые веса.</param>
public static void UpdateWeights(this NeuronModel neuronModel, List <double> weights)
{
if (neuronModel.Weights.Count != weights.Count)
{
var exception = new Exception("Несоответствие по количеству весов.");
Console.WriteLine(BL.Constants.ConsoleMessageConstants.ERROR_MESSAGE +
exception.ToString());
Console.ReadKey();
return;
}
neuronModel.LastWeights = neuronModel.Weights;
neuronModel.Weights = weights;
}
public void PickObject(int x, int y)
{
if (m_simulationModel == null)
{
return;
}
m_pickRay = PickRay.Pick(x, y, m_camera, m_control.Size, ProjectionMatrix);
NeuronModel neuron = FindFirstNeuron(m_pickRay, m_simulationModel.Regions);
if (neuron != null)
{
ToggleNeuron(neuron);
}
}
public void OpenObserver(NeuronModel neuron, Scene scene)
{
ObserverDefinition definition = CreateObserverDefinition(neuron);
// TODO(HonzaS): Factory + injection.
var observer = new CanvasObserver(definition, Conductor.ModelProvider);
observer.Log = Log;
var form = new ObserverForm(this, observer);
form.Text = $"Neuron {neuron.Index}, region {neuron.RegionModel.Index} - '{definition.Type}'";
var handle = new ObserverHandle(observer, form, scene);
Observers.Add(handle);
form.Show();
RefreshObserverRequests();
}
private void SetupRegionWithSynapse(out RegionModel addedRegion, out SynapseModel addedSynapse)
{
addedRegion = SetupRegion(m_applier, m_model, 1);
var addedNeuron1 = new NeuronModel(1, "neuronType", addedRegion, Vector3.One);
var addedNeuron2 = new NeuronModel(2, "neuronType", addedRegion, Vector3.UnitY);
addedSynapse = new SynapseModel(addedRegion, addedNeuron1, addedRegion, addedNeuron2);
ResponseMessage diff = ModelResponseBuilder.Build(
addedNeurons: new List <NeuronModel> {
addedNeuron1, addedNeuron2
},
addedSynapses: new List <SynapseModel> {
addedSynapse
});
ApplyModelDiff(diff);
}
private NeuronModel FindFirstNeuron(PickRay pickRay, IEnumerable <RegionModel> regions)
{
float closestDistance = float.MaxValue;
NeuronModel closestNeuron = null;
foreach (RegionModel region in regions)
{
foreach (NeuronModel neuron in region.Neurons)
{
float distance = neuron.DistanceToRayOrigin(pickRay);
if (distance < closestDistance)
{
closestNeuron = neuron;
closestDistance = distance;
}
}
}
return(closestNeuron);
}
/// <summary>
/// Инициализация слоя.
/// </summary>
public void Initilize()
{
var inputs = new List <double>();
var weights = new List <double>();
var emptyLastWeights = new List <double>();
foreach (var neuron in _hiddenLayerData)
{
inputs.Add(neuron.Output);
weights.Add(GetInitializedWeight());
emptyLastWeights.Add(0d);
}
_outputNeuron = new NeuronModel(inputs, weights)
{
LastWeights = emptyLastWeights
};
}
/// <summary>
/// Инициализировать выходной слой.
/// </summary>
public override void Initialize()
{
var weights = new List <double>();
var emptyWeights = new List <double>();
var random = new Random();
for (var wIndex = 0; wIndex < _inputData.Count; ++wIndex)
{
weights.Add(random.NextDouble());
emptyWeights.Add(0d);
System.Threading.Thread.Sleep(10);
}
_outputNeuron = new NeuronModel(_inputData, weights)
{
LastWeights = emptyWeights
};
_output = _outputNeuron.Output;
}
/// <summary>
/// Приготовить к распознаванию.
/// </summary>
/// <param name="neuronIndexToWeightsValueDictionary">Словарь значений весов,
/// где ключ - индекс нейрона, значение - веса нейрона.</param>
public void RecognizeMode(Dictionary <int, List <double> > neuronIndexToWeightsValueDictionary)
{
_hiddenLayerData = new List <NeuronModel>();
var countOfNeuronsInHiddenLayer = (int)_convolutionalLayerData.Count / 2;
var inputs = new List <double>();
_convolutionalLayerData.ForEach(neuronOfConvolutionalLayer
=> inputs.Add(neuronOfConvolutionalLayer.Output));
for (var index = 0; index < countOfNeuronsInHiddenLayer; ++index)
{
if (!neuronIndexToWeightsValueDictionary.TryGetValue(index, out var weights))
{
ErrorHelper.GetDataError();
}
var neuron = new NeuronModel(inputs, weights);
_hiddenLayerData.Add(neuron);
}
}
/// <summary>
/// Приготовить к распознаванию.
/// </summary>
/// <param name="filterCore">Ядро фильтра.</param>
public void RecognizeMode(double[,] filterCore)
{
_convolutionalLayerData = new List <NeuronModel>();
var offset = MatrixConstants.MATRIX_SIZE - MatrixConstants.FILTER_MATRIX_SIZE;
var step = MatrixConstants.MATRIX_SIZE - offset;
for (var xIndex = 0; xIndex < step; ++xIndex)
{
for (var yIndex = 0; yIndex < step; ++yIndex)
{
var inputs = new List <double>();
var weights = new List <double>();
GetDataByFilterCore(filterCore, xIndex, yIndex, inputs, weights);
var neuron = new NeuronModel(inputs, weights);
_convolutionalLayerData.Add(neuron);
}
}
}
/// <summary>
/// Создать нейрон по имеющимся данным.
/// </summary>
/// <returns>Возвращает нейрон.</returns>
private NeuronModel CreateNeuronByData()
{
var inputs = new List <double>();
var weights = new List <double>();
var emptyLastWeights = new List <double>();
foreach (var neuron in _convolutionalLayerData)
{
inputs.Add(neuron.Output);
weights.Add(GetInitializedWeight());
emptyLastWeights.Add(0d);
}
var hiddenLayerNeuron = new NeuronModel(inputs, weights)
{
LastWeights = emptyLastWeights
};
return(hiddenLayerNeuron);
}
/// <summary>
/// Инициализация слоя.
/// </summary>
public override void Initialize()
{
_neurons = new List <NeuronModel>();
_output = new List <double>();
if (_countOfNeurons.Equals(0))
{
_output = _inputData;
return;
}
var neurons = new List <NeuronModel>();
for (var index = 0; index < _countOfNeurons; ++index)
{
var weights = new List <double>();
var emptyWeights = new List <double>();
var random = new Random();
for (var wIndex = 0; wIndex < _inputData.Count; ++wIndex)
{
weights.Add(random.NextDouble());
emptyWeights.Add(0d);
System.Threading.Thread.Sleep(10);
}
var neuron = new NeuronModel(_inputData, weights)
{
LastWeights = emptyWeights
};
neurons.Add(neuron);
}
neurons.ForEach(neuron => _output.Add(neuron.Output));
_neurons.AddRange(neurons);
}
/// <summary>
/// Инициализировать для прогноза.
/// </summary>
/// <param name="weightsDictionary">Словарь весов.</param>
public void SwitchToForeCast(Dictionary <int, List <double> > weightsDictionary)
{
_neurons = new List <NeuronModel>();
_output = new List <double>();
if (_countOfNeurons.Equals(0))
{
_output = _inputData;
return;
}
var neurons = new List <NeuronModel>();
for (var index = 0; index < _countOfNeurons; ++index)
{
var weights = weightsDictionary[index];
var neuron = new NeuronModel(_inputData, weights);
neurons.Add(neuron);
}
_neurons.AddRange(neurons);
_neurons.ForEach(neuron => _output.Add(neuron.Output));
}
/// <summary>
/// Обновление данных.
/// </summary>
/// <param name="dataSet">Датасет.</param>
public void UpdateData(List <double> dataSet)
{
if (_countOfNeurons.Equals(0))
{
_output = dataSet;
return;
}
var updatedNeurons = new List <NeuronModel>();
foreach (var neuron in _neurons)
{
var newNeuron = new NeuronModel(dataSet, neuron.Weights)
{
LastWeights = neuron.LastWeights
};
updatedNeurons.Add(neuron);
}
_neurons = updatedNeurons;
_output.Clear();
_neurons.ForEach(neuron => _output.Add(neuron.Output));
}
/// <summary>
/// Расчёт дельты для нейронов скрытого слоя.
/// </summary>
/// <param name="neuronOutput">Вывод нейрона.</param>
/// <param name="indexInLayer">Индекс в скрытом слое.</param>
/// <param name="outputNeuron">Нейрон выходного слоя.</param>
/// <returns></returns>
private double GetHiddenLayerNeuronDelta(
double neuronOutput, int indexInLayer, NeuronModel outputNeuron) =>
DerivativeActivationFunction(neuronOutput) *
outputNeuron.Inputs[indexInLayer] * outputNeuron.Delta;
private bool TryFindNeuron(CompositeNeuronId neuronId, out NeuronModel neuron)
{
return(TryFindNeuron(neuronId.RegionIndex, neuronId.NeuronIndex, out neuron));
}
private ObserverDefinition CreateObserverDefinition(NeuronModel neuron)
{
var definition = new ObserverDefinition(neuron.Index, neuron.RegionModel.Index, m_observerType);
return(definition);
}
public void CloseObserver(NeuronModel neuron)
{
ObserverDefinition definition = CreateObserverDefinition(neuron);
CloseObserver(definition);
}
/// <summary>
/// Инициализировать для прогноза.
/// </summary>
/// <param name="weightsDictionary">Веса.</param>
public void SwitchToForeCast(List <double> weights)
{
_outputNeuron = new NeuronModel(_inputData, weights);
_output = _outputNeuron.Output;
}