mlpmath.h

#ifndef MLPMATH
#define MLPMATH

#include "types.h"
#include <cmath>

inline double norm(const EigenMatrix &input)
{
    return (input.transpose()*input).trace()/input.cols();
}

inline double norm2(const EigenMatrix &input)
{
    return input.array().square().sum();
}

inline double sigmoid(double const &value) {
    return 1/(1+exp(-value));
}

inline double sigmoidDerivative(double const &value) {
    return sigmoid(value)*(1-sigmoid(value));
}

inline double sigmoidDerivativeA(double const &value) {
    return value*(1-value);
}

inline double tanH(double const &value) {
    return 1.7159*tanh(value*2/3);
}

inline double tanHDerivative(double const &value) {
    return 1.7159*2/3*(1-pow(tanh(value*2/3),2));
}

inline double chooseConstFunction(double (*function)(double const &), double value)
{
    return (*function)(value) ;
}

inline double chooseFunction(double (*function)(double), double value)
{
    return (*function)(value) ;
}

inline EigenMatrix addBias(const EigenMatrix &matrix)
{
    EigenMatrix newMatrix(matrix.rows()+1, matrix.cols());
    newMatrix.block(0,0, matrix.rows(), matrix.cols()) = matrix;
    newMatrix.block(matrix.rows(),0, 1, matrix.cols()) = -EigenVector::Ones(matrix.cols()).transpose();
    return newMatrix;
}

inline EigenMatrix activation(const EigenMatrix &weights, const EigenMatrix &inputVector)
{
    return (weights * addBias(inputVector)).unaryExpr(ptr_fun(sigmoid));
}

inline EigenMatrix activation(const EigenMatrix &weights, const EigenMatrix &inputVector, double (*function)(double const &))
{
    return (weights * addBias(inputVector)).unaryExpr(ptr_fun(function));
}

#endif // MLPMATH