Class mcmc_stepper_rw (o2scl)¶
-
template<class func_t, class data_t, class vec_t>
class mcmc_stepper_rw : public o2scl::mcmc_stepper_base<func_t, data_t, vec_t>¶ A simple random-walk stepper for MCMC.
This stepper performs a random walk. Given the parameter \( p_i \), the lower limit \( \ell_i \), the upper limit \( u_i \), a random number \( r_i \in [0,1) \), and the “step factor” \( s_i \), the new coordinate \( p_{\mathrm{new,i}} \) is
\[ p_{\mathrm{new,i}} = p_i + \frac{(2 r_i-1)}{s_i}(u_i - \ell_i) \]The value of \( s_i \) is taken fromThus larger values of step_fac result in smaller steps.step_fac[k % step_fac.size()]
If the final point in parameter space, \( p_{\mathrm{new}} \), is out of bounds, then the value of
func_retis set to mcmc_stepper_base::mcmc_skip (which will lead to a rejection in mcmc_para or its children).Public Functions
-
inline virtual const char *step_type()¶
Stepper type, “RW”.
-
inline mcmc_stepper_rw()¶
-
inline virtual ~mcmc_stepper_rw()¶
-
inline virtual void write_params(o2scl_hdf::hdf_file &hf)¶
Write stepper parameters to the HDF5 file.
-
inline virtual void step(size_t i_thread, size_t n_params, func_t &f, vec_t ¤t, vec_t &next, double w_current, double &w_next, vec_t &low, vec_t &high, int &func_ret, bool &accept, data_t &dat, rng<> &r, int verbose)¶
Construct a step.
This function constructs
nextandw_next, the next point and log weight in parameter space. The objective functionfis then evaluated at the new point, the return value is placed infunc_ret, and the step acceptance or rejection is stored inaccept.
-
inline virtual const char *step_type()¶