AlgoParamsDEVI {DEBBI}

AlgoParamsDEVI

Description

get control parameters for DEVI function

Usage

AlgoParamsDEVI(
  n_params,
  param_names = NULL,
  n_chains = NULL,
  n_iter = 1000,
  init_sd = 0.01,
  init_center = 0,
  n_cores_use = 1,
  step_size = NULL,
  jitter_size = 1e-06,
  parallel_type = "none",
  use_QMC = TRUE,
  purify = NULL,
  quasi_rand_seq = "halton",
  n_samples_ELBO = 10,
  LRVB_correction = TRUE,
  n_samples_LRVB = 25,
  neg_inf = -750,
  thin = 1,
  burnin = 0,
  return_trace = FALSE,
  crossover_rate = 1
)

Arguments

n_params

number of free parameters estimated

param_names

optional vector of parameter names

n_chains

number of particle chains used for optimization, 3*n_params is the default value

n_iter

number of iterations to run the sampling algorithm, 1000 is default

init_sd

positive scalar or n_params-dimensional numeric vector, determines the standard deviation of the Gaussian initialization distribution

init_center

scalar or n_params-dimensional numeric vector, determines the mean of the Gaussian initialization distribution

n_cores_use

number of cores used when using parallelization.

step_size

positive scalar, jump size in DE crossover step, default is 2.38/sqrt(2*n_params).

jitter_size

positive scalar, noise is added during crossover step from Uniform(-jitter_size,jitter_size) distribution. 1e-6 is the default value.

parallel_type

string specifying parallelization type. 'none','FORK', or 'PSOCK' are valid values. 'none' is default value.

use_QMC

logical, if true, a quasi-Monte Carlo estimator is used to estimate ELBO during optimization. default is TRUE.

purify

an integer, every 'purify'-th iteration, the Monte Carlo estimator of the ELBO is recalculated. This can help deal with noisy and outlier estimates of the ELBO. Default value is 25. If use_QMC is TRUE, purification is disabled as it is redundant.

quasi_rand_seq

type of low discrepancy sequence used for quasi Monte Carlo integration, either 'sobol' or 'halton'. LRVB correction always use QMC. Default is 'sobol'.

n_samples_ELBO

number of samples used for the Monte Carlo estimator of the ELBO (the objective function). default is 10.

LRVB_correction

logical, if true, LRVB covariance correction (Giordano, Brodderick, & Jordan 2018; Galdo, Bahg, & Turner 2020) is attempted.

n_samples_LRVB

number of samples used for LRVB correction. default is 25.

neg_inf

if density for a given value of theta is numerically 0 for q, this value is assigned for log density. This helps with numeric stability of algorithm. Default value is -750.

thin

positive integer, only every 'thin'-th iteration will be stored. Default value is 1. Increasing thin will reduce the memory required, while running algorithm for longer.

burnin

number of initial iterations to discard. Default value is 0.

return_trace

logical, if true, function returns particle trajectories. This is helpful for diagnosing convergence or debugging model code. Function will return an iteration/thin $x$ n_chains $x$ n_params array and the estimated ELBO of each particle in a iteration/thin x n_chains array.

crossover_rate

number on the interval (0,1]. Determines the probability a parameter on a chain is updated on a given crossover step, sampled from a Bernoulli distribution.

Value

list of control parameters for the DEVI function