autoppl
v0.8
A C++ template library for probabilistic programming
|
#include <step_adapter.hpp>
Public Member Functions | |
StepAdapter (double _log_eps) | |
void | init (double _log_eps) |
void | adapt (double alpha_ratio) |
void | reset () |
Public Attributes | |
size_t | counter = 0 |
double | log_eps = 0. |
double | log_eps_bar = 0. |
double | H_bar = 0. |
double | mu = 0. |
const double | mu_constant = std::log(10) |
StepConfig | step_config |
Adaptive values in NUTS algorithm using dual-averaging. Currently, only log_eps, log_eps_bar, and H_bar are adaptive.
|
inline |
Constructs a step adapter with the given log epsilon. Initializes mu to log(10*_log_eps).
|
inline |
Adapts log_eps, log_eps_bar, and H_bar.
|
inline |
Must be called before beginning to adapt.
|
inline |
Reset all variables that change per iteration
size_t ppl::mcmc::StepAdapter::counter = 0 |
double ppl::mcmc::StepAdapter::H_bar = 0. |
double ppl::mcmc::StepAdapter::log_eps = 0. |
double ppl::mcmc::StepAdapter::log_eps_bar = 0. |
double ppl::mcmc::StepAdapter::mu = 0. |
const double ppl::mcmc::StepAdapter::mu_constant = std::log(10) |
StepConfig ppl::mcmc::StepAdapter::step_config |