Iterative updating of beta distribution. More...

#include <bsta/bsta_beta.h>
#include <bsta/bsta_attributes.h>
#include <vcl_algorithm.h>
#include <vcl_iostream.h>

Classes
struct	bsta_beta_fitness< beta_ >
class	bsta_beta_updater< beta_ >
	An updater for statistically updating beta distributions. More...
class	bsta_mix_beta_updater< mix_dist_ >
Functions
template<class T >
void	bsta_update_beta (bsta_beta< T > &beta_dist, T rho, const T &sample)
	Update the statistics given a 1D beta distribution and a learning rate.
template<class T >
void	bsta_update_beta (bsta_beta< T > &beta_dist, T rho, const T &sample, const T &min_var)

Detailed Description

Iterative updating of beta distribution.

In this implementation $\alpha>=1$ and $\beta>=1$ . In order to ensure this $\mu(\mu(1-\mu)/var-1)>1$ and $(1-\mu)(\mu(1-\mu)/var-1)>1$

The distance of beta distribution is given as $$ -(\alpha-1)log(x/\mu)-(\beta-1)\log((1-x)/(1-\mu)) > 3$ $

    Modifications

Definition in file bsta_beta_updater.h.

Function Documentation

template<class T >

Update the statistics given a 1D beta distribution and a learning rate.

Note:: if rho = 1/(num observations) then this is just an online cumulative average

Definition at line 30 of file bsta_beta_updater.h.

template<class T >

Definition at line 63 of file bsta_beta_updater.h.