mssm (version 0.1.6)

mssm_control: Auxiliary for Controlling Multivariate State Space Model Fitting

Description

Auxiliary function for mssm.

Usage

mssm_control(N_part = 1000L, n_threads = 1L, covar_fac = 1.2,
  ftol_rel = 1e-06, nu = 8, what = "log_density",
  which_sampler = "mode_aprx", which_ll_cp = "no_aprx", seed = 1L,
  KD_N_max = 10L, aprx_eps = 0.001, ftol_abs = 1e-04,
  ftol_abs_inner = 1e-04, la_ftol_rel = -1, la_ftol_rel_inner = -1,
  maxeval = 10000L, maxeval_inner = 10000L, use_antithetic = FALSE)

Arguments

N_part

integer greater than zero for the number of particles to use.

n_threads

integer greater than zero for the number of threads to use.

covar_fac

positive numeric scalar used to scale the covariance matrix in the proposal distribution.

ftol_rel

positive numeric scalar with convergence threshold passed to nloptr if the mode approximation method is used for the proposal distribution.

nu

degrees of freedom to use for the multivariate \(t\)-distribution that is used as the proposal distribution. A multivariate normal distribution is used if nu <= 2.

what

character indicating what to approximate. "log_density" implies only the log-likelihood. "gradient" also yields a gradient approximation. "Hessian" also yields an approximation of the observed information matrix.

which_sampler

character indicating what type of proposal distribution to use. "mode_aprx" yields a Taylor approximation at the mode. "bootstrap" yields a proposal distribution similar to the common bootstrap filter.

which_ll_cp

character indicating what type of computation should be performed in each iteration of the particle filter. "no_aprx" yields no approximation. "KD" yields an approximation using a dual k-d tree method.

seed

integer with seed to pass to set.seed.

KD_N_max

integer greater than zero with the maximum number of particles to include in each leaf of the two k-d trees if the dual k-d trees method is used.

aprx_eps

positive numeric scalar with the maximum error if the dual k-d tree method is used.

ftol_abs, ftol_abs_inner, la_ftol_rel, la_ftol_rel_inner, maxeval, maxeval_inner

scalars passed to nlopt when estimating parameters with a Laplace approximation. The _inner denotes the values passed in the inner mode estimation. The mode estimation is done with a custom Newton<U+2013>Raphson method

use_antithetic

logical which is true if antithetic variables should be used.

See Also

mssm.

See the README of the package for details of the dual k-d tree method at https://github.com/boennecd/mssm.

Examples

# NOT RUN {
library(mssm)
str(mssm_control())
str(mssm_control(N_part = 2000L))
# }