Params

template<class Mag>
class Params

Class to wrap training parameters.

This class is used by the rFBP functions to facilitate the moving of a set of training parameters along the series of functions.

Template Parameters
  • Mag: magnetization used

Public Functions

Params(const int &max_iters, const double &damping, const double &epsil, const double &beta, const double &r, const double &gamma, const std::string &accuracy1, const std::string &accuracy2)

Parameter constructor.

Note
In the constructor the value of gamma is converted to the appropriated Mag type.
Parameters
  • max_iters: Number of iterations
  • damping: Damping factor
  • epsil: Error tollerance
  • beta: 1 / KT
  • r: Number of replicas - 1
  • gamma: Hyperbolic tangent of distance weight between replicas (γ) as Mag object
  • accuracy1: Updating accuracy of cavity probability (messages of hidden layers)
  • accuracy2: Updating accuracy of cavity probability (messages of otuput node)

~Params()

Default destructor.

Public Members

long int max_iters

Number of iterations.

double damping

Damping factor.

double epsil

Error tollerance.

double beta

1/kT

double r

Number of replicas -1.

Mag tan_gamma

Hyperbolic tangent of distance weight between replicas (γ)

std::string accuracy1

Updating accuracy of cavity probability (messages of hidden layers)

std::string accuracy2

Updating accuracy of cavity probability (messages of output node)