internal InputLayer2D(Size dimensions, Layer2D prevLayer, int sliceCount, string function = null) : base(prevLayer, null, sliceCount)
{
this.inputs = new MemoryBuffer2D <float> [sliceCount];
var widthOutput = (int)dimensions.Width;
var heightOutput = (int)dimensions.Height;
for (int i = 0; i < sliceCount; i++)
{
this.inputs[i] = GPUHelper.CreateBuffer(widthOutput, heightOutput);
this.activated[i] = GPUHelper.CreateBuffer(widthOutput, heightOutput);
this.sumInput[i] = GPUHelper.CreateBuffer(widthOutput, heightOutput);
}
if (function != null)
{
MethodInfo methodInfo = Type.GetType("DeepDave.Layer.Kernels.NormalizationFunctions").GetMethod(function, BindingFlags.NonPublic | BindingFlags.Static);
normalize = GPUHelper.CreateKernel(methodInfo).CreateLauncherDelegate <Action <AcceleratorStream, Index2, ArrayView2D <float>, ArrayView2D <float>, ArrayView <float> > >();
}
}
public FullyConnectedLayer2D(Size outputSize, int sliceCount, Layer2D prevLayer, string activationFunction) : base(prevLayer, activationFunction, sliceCount)
{
var x = (int)outputSize.Width;
var y = (int)outputSize.Height;
var fac = GetSuitableFactorForFunction(function, x * y);
for (int i = 0; i < sliceCount; i++)
{
float[] source = { Config.learningRate, fac };
this.variable[i] = GPUHelper.CreateBuffer(source, 2);
this.bias[i] = GPUHelper.CreateBuffer(x, y);
this.activated[i] = GPUHelper.CreateBuffer(x, y);
this.sumInput[i] = GPUHelper.CreateBuffer(x, y);
this.error[i] = GPUHelper.CreateBuffer(x, y);
this.weight[i] = GPUHelper.CreateBuffer(x, y, this.GetWeightCount());
this.derived[i] = GPUHelper.CreateBuffer(x, y);
}
}
public ConvolutionalLayer2D(int radius, int sliceCount, Layer2D prevLayer, string activationFunction) : base(prevLayer, activationFunction, sliceCount)
{
this.radius = radius;
var x = (int)prevLayer.GetActivatedBuffer(0).Extent.X;
var y = (int)prevLayer.GetActivatedBuffer(0).Extent.Y;
this.fac = GetSuitableFactorForFunction(function, (2 * radius + 1) * (2 * radius + 1));
for (int i = 0; i < sliceCount; i++)
{
float[] source = { Config.learningRate, fac, radius };
this.variable[i] = GPUHelper.CreateBuffer(source, source.Length);
this.bias[i] = GPUHelper.CreateBuffer(x, y);
this.activated[i] = GPUHelper.CreateBuffer(x, y);
this.sumInput[i] = GPUHelper.CreateBuffer(x, y);
this.error[i] = GPUHelper.CreateBuffer(x, y);
this.weight[i] = GPUHelper.CreateBuffer(x, y, this.GetWeightCount());
this.derived[i] = GPUHelper.CreateBuffer(x, y);
}
}
public RandomLayer2D(Layer2D prevLayer, int sliceCount, int maxConnectionsPerNeuron) : base(prevLayer, null, sliceCount)
{
var x = (int)prevLayer.GetActivatedBuffer(0).Extent.X;
var y = (int)prevLayer.GetActivatedBuffer(0).Extent.Y;
var fac = GetSuitableFactorForFunction(function, x * y);
var weightFactory = new RandomMatrixFactory(x, y, maxConnectionsPerNeuron, RandomMatrixFactory.GenerationType.Float);
var conFactory = new RandomMatrixFactory(x, y, maxConnectionsPerNeuron * 2, RandomMatrixFactory.GenerationType.Integer, maxConnectionsPerNeuron);
connectionInfo = new MemoryBuffer3D <int> [sliceCount];
newConnectionInfo = new MemoryBuffer3D <int> [sliceCount];
for (int i = 0; i < sliceCount; i++)
{
float[] source = { Config.learningRate, fac };
this.variable[i] = GPUHelper.CreateBuffer(source, 2);
this.bias[i] = GPUHelper.CreateBuffer(x, y);
this.activated[i] = GPUHelper.CreateBuffer(x, y);
this.sumInput[i] = GPUHelper.CreateBuffer(x, y);
this.error[i] = GPUHelper.CreateBuffer(x, y);
this.weight[i] = GPUHelper.CreateBuffer(x, y, maxConnectionsPerNeuron);
this.derived[i] = GPUHelper.CreateBuffer(x, y);
bool b = true;
while (b)
{
if (conFactory.intQueue3D.Count > 1)
{
b = false;
}
}
int[,,] arr;
conFactory.intQueue3D.TryDequeue(out arr);
this.connectionInfo[i] = GPUHelper.CreateBuffer(arr, x, y, maxConnectionsPerNeuron * 2);
conFactory.intQueue3D.TryDequeue(out arr);
this.newConnectionInfo[i] = GPUHelper.CreateBuffer(arr, x, y, maxConnectionsPerNeuron * 2);
}
var extent = connectionInfo[0].Extent;
}