/// <summary>
/// Computes activation function and its derivative
/// </summary>
/// <param name="n">int</param>
/// <param name="net">double - net</param>
/// <param name="acct">VectorHorizontal - activation matrix</param>
/// <param name="gain">VectorHorizontal - gain matrix</param>
/// <returns>FunctionRD - function result</returns>
/// <remarks>Remember that values are passed by ref to speed up</remarks>
public static FunctionRD computeFunctionDervative(ref int n, ref double net, ref Activation acct, ref Gain gain)
{
FunctionChoice function = (FunctionChoice)(int)acct.Data[0][n];
FunctionRD rd = new FunctionRD();
switch (function)
{
case FunctionChoice.LinearNeuron: //for output layer
{
rd.FunctionResult = gain.Data[0][n] * net;
rd.FunctionDerivative = gain.Data[0][n];
} break;
case FunctionChoice.UnipolarNeuron:
{
rd.FunctionResult = 1 / (1 + System.Math.Exp(-gain.Data[0][n] * net));
rd.FunctionDerivative = gain.Data[0][n] * (1 - rd.FunctionResult) * rd.FunctionResult;
} break;
case FunctionChoice.BipolarNeuron: //for hidden layer
{
rd.FunctionResult = System.Math.Tanh(gain.Data[0][n] * net);
rd.FunctionDerivative = gain.Data[0][n] * (1 - System.Math.Pow(rd.FunctionResult, 2));
} break;
case FunctionChoice.BipolarElliotNeuron:
{
rd.FunctionResult = gain.Data[0][n] * net / (1 + gain.Data[0][n] * System.Math.Abs(net));
rd.FunctionDerivative = 1 / System.Math.Pow((gain.Data[0][n] * System.Math.Abs(net) + 1), 2);
} break;
case FunctionChoice.UnipolarElliotNeuron:
{
rd.FunctionResult = 2 * gain.Data[0][n] * net / (1 + gain.Data[0][n] * System.Math.Abs(net)) - 1;
rd.FunctionDerivative = 2 * gain.Data[0][n] / System.Math.Pow((gain.Data[0][n] * System.Math.Abs(net) + 1), 2);
} break;
default:
{
throw new NeuralNetworkError(Properties.Settings.Default.FE1);
}
}
return(rd);
}
/// <summary>
/// Computes activation function and its derivative
/// </summary>
/// <param name="n">int</param>
/// <param name="net">double - net</param>
/// <param name="acct">VectorHorizontal - activation matrix</param>
/// <param name="gain">VectorHorizontal - gain matrix</param>
/// <returns>FunctionRD - function result</returns>
/// <remarks>Remember that values are passed by ref to speed up</remarks>
public static FunctionRD computeFunctionDervative(ref int n, ref double net, ref Activation acct, ref Gain gain)
{
FunctionChoice function = (FunctionChoice)(int)acct.Data[0][n];
FunctionRD rd = new FunctionRD();
switch (function)
{
case FunctionChoice.LinearNeuron://for output layer
{
rd.FunctionResult = gain.Data[0][n] * net;
rd.FunctionDerivative = gain.Data[0][n];
}break;
case FunctionChoice.UnipolarNeuron:
{
rd.FunctionResult = 1 / (1 + System.Math.Exp(-gain.Data[0][n] * net));
rd.FunctionDerivative = gain.Data[0][n] * (1 - rd.FunctionResult) * rd.FunctionResult;
}break;
case FunctionChoice.BipolarNeuron://for hidden layer
{
rd.FunctionResult = System.Math.Tanh(gain.Data[0][n] * net);
rd.FunctionDerivative = gain.Data[0][n] * (1 - System.Math.Pow(rd.FunctionResult,2));
}break;
case FunctionChoice.BipolarElliotNeuron:
{
rd.FunctionResult = gain.Data[0][n] * net / (1 + gain.Data[0][n] * System.Math.Abs(net));
rd.FunctionDerivative = 1 / System.Math.Pow((gain.Data[0][n] * System.Math.Abs(net) + 1), 2);
}break;
case FunctionChoice.UnipolarElliotNeuron:
{
rd.FunctionResult = 2 * gain.Data[0][n] * net / (1 + gain.Data[0][n] * System.Math.Abs(net)) - 1;
rd.FunctionDerivative = 2 * gain.Data[0][n] / System.Math.Pow((gain.Data[0][n] * System.Math.Abs(net) + 1), 2);
}break;
default:
{
throw new NeuralNetworkError(Properties.Settings.Default.FE1);
}
}
return rd;
}
