/// <summary>
/// 构造函数
/// </summary>
/// <param name="inputWidth"></param>
/// <param name="inputHeight"></param>
/// <param name="receptiveFieldWidth"></param>
/// <param name="receptiveFieldHeight"></param>
/// <param name="offsetWidth"></param>
/// <param name="offsetHeight"></param>
/// <param name="activationFunctionType">激活函数类型，1:tanh,2:PReLU,3:Sigmoid</param>
/// <param name="inputCount"></param>
/// <param name="outputCount"></param>
public CnnKernel(int inputWidth, int inputHeight, int receptiveFieldWidth, int receptiveFieldHeight,
int offsetWidth, int offsetHeight, ActivationFunctionTypes activationFunctionType, int inputCount, int outputCount, bool standardization)
{
this.receptiveFieldWidth = receptiveFieldWidth;
this.receptiveFieldHeight = receptiveFieldHeight;
this.inputWidth = inputWidth;
this.inputHeight = inputHeight;
this.offsetWidth = offsetWidth;
this.offsetHeight = offsetHeight;
this.ActivationFunctionType = activationFunctionType;
this.InputCount = inputCount;
this.OutputCount = outputCount;
this.Standardization = standardization;
this.ConvolutionKernelWidth = Convert.ToInt32(Math.Ceiling((inputWidth - receptiveFieldWidth) / (double)offsetWidth)) + 1;
this.ConvolutionKernelHeight = Convert.ToInt32(Math.Ceiling((inputHeight - receptiveFieldHeight) / (double)offsetHeight)) + 1;
ShareWeight = new List <double[, ]>();
meanDeltaWeight = new List <double[, ]>();
for (int i = 0; i < inputCount; i++)
{
ShareWeight.Add(new double[receptiveFieldWidth, receptiveFieldHeight]);
meanDeltaWeight.Add(new double[receptiveFieldWidth, receptiveFieldHeight]);
}
meanListDeltaWeight = new List <List <double[, ]> >();
meanListDeltaOffset = new List <double>();
InitShareWeight();
}
/// <summary>
/// 构造函数
/// </summary>
/// <param name="InputCount"></param>
/// <param name="OutputCount"></param>
/// <param name="activationFunctionType">激活函数类型，1:tanh,2:PReLU,3:Sigmoid</param>
public CnnFullLayer(int InputCount, int OutputCount, ActivationFunctionTypes activationFunctionType, bool standardization)
{
this.InputCount = InputCount;
this.OutputCount = OutputCount;
this.ActivationFunctionType = activationFunctionType;
this.Standardization = standardization;
InputWeight = new double[OutputCount, InputCount];
OutputOffset = new double[OutputCount];
meanDeltaWeight = new double[OutputCount, InputCount];
meanDeltaOffset = new double[OutputCount];
meanListDeltaWeight = new List <double[, ]>();
meanListDeltaOffset = new List <double[]>();
InitInputWeight();
}
//Called from the neural net, sets the outputs based on node's connections
//Used for the hidden layers and final output layer
public void SetOutput(ActivationFunctionTypes activationFunctionUsed)
{
output = 0;
//Cycle through all the inputs into this node
foreach (var nodeConnection in inputConnections)
{
//Add them together, including the weighting for each individual node
SumInput(nodeConnection);
}
//Add the node's bias on to outputs
output += bias;
//Adjust the data so we receive sensible results
ActivationTanH(output);
}
/// <summary>
/// 构造函数
/// </summary>
/// <param name="inputWidth"></param>
/// <param name="inputHeight"></param>
/// <param name="receptiveFieldWidth"></param>
/// <param name="receptiveFieldHeight"></param>
/// <param name="poolingType">池化类型，1:平均池化(Mean Pooling),2:最大值池化(Max Pooling)</param>
public CnnPooling(int inputWidth, int inputHeight, int receptiveFieldWidth, int receptiveFieldHeight,
ActivationFunctionTypes activationFunctionType, PoolingTypes poolingType, int inputCount, int outputCount)
{
this.receptiveFieldWidth = receptiveFieldWidth;
this.receptiveFieldHeight = receptiveFieldHeight;
this.inputWidth = inputWidth;
this.inputHeight = inputHeight;
this.ActivationFunctionType = activationFunctionType;
this.PoolingType = poolingType;
this.InputCount = inputCount;
this.OutputCount = outputCount;
this.ConvolutionKernelWidth = Convert.ToInt32(Math.Ceiling(inputWidth / (double)receptiveFieldWidth));
this.ConvolutionKernelHeight = Convert.ToInt32(Math.Ceiling(inputHeight / (double)receptiveFieldWidth));
OutputPoolingMax = new int[ConvolutionKernelWidth, ConvolutionKernelHeight];
InitShareWeight();
}
