Get list of mcmcglms run across values of slice sampling tuning parameters

The function simply performs an lapply over mcmcglm for the values provided of tuning parameter values. See general usage in the vignette

Usage

mcmcglm_across_tuningparams(
  ...,
  tuning_parameter_name = "w",
  formula,
  family,
  data,
  beta_prior = distributional::dist_normal(0, 1),
  log_likelihood_extra_args = list(sd = 1),
  sample_method = c("slice_sampling", "normal-normal"),
  qslice_fun = qslice::slice_stepping_out,
  n_samples = 500,
  burnin = 100,
  parallelise = FALSE,
  n_cores = as.numeric(Sys.getenv("NUMBER_OF_PROCESSORS")) - 1
)

Arguments

...: A list or vector with values of the tuning parameter
tuning_parameter_name: The name of the tuning parameter. Fx. for the default qslice_fun qslice::slice_stepping_out, the tuning parameter is called w
formula: an object of class "formula" (or one that can be coerced to that class): a symbolic description of the model to be fitted. See more details at stats::glm
family: a description of the error distribution and link function to be used in the model. This can be a character string naming a family function, a family function or the result of a call to a family function. (See family for details of family functions.)
data: an optional data frame, list or environment (or object coercible by as.data.frame to a data frame) containing the variables in the model. If not found in data, the variables are taken from environment(formula), typically the environment from which the function is called.
beta_prior: a distribution object created by a function from the distributional package. Could fx. be distributional::dist_normal(mean = 0, sd = 1).
log_likelihood_extra_args: a named list with arguments passed onto the log_density function. Fx. specification of log_likelihood_extra_args = list(sd = x) is needed for the case of family = "gaussian"
sample_method: a character specifying the method used for sampling. The default "slice_sampling" is the intended value in most cases, as it works for any specification of family and beta_prior. "normal-normal" uses a conditional normal distribution to sample from in case of conjugate prior with gaussian response and beta_prior. Implemented for testing purposes but works for that niche case.
qslice_fun: a function from the qslice package. Default is qslice::slice_stepping_out which uses the slice sampler from Neal 2003, but all functions are available.
n_samples: a numeric with number of samples to draw of each parameter(/variable) in the model
burnin: a numeric with the number of samples to be marked as "burnin". Burnin samples are not included in the beta_mean calculation to increase finite sample performance of the LLN estimate
parallelise: a logical if the runs of the algorithm across values of n_vars should be parallelised using future.apply::future_lapply
n_cores: a numeric with the number of cores to use for parallelisation. Default is 1 less than the number of available cores

Examples

# Create test data for different scenarios
n <- 100
x1 <- rnorm (n)
x2 <- rbinom (n, 1, .5)
b0 <- 1
b1 <- 1.5
b2 <- 2
lin_pred <- b0+b1*x1+b2*x2

#############################################
# Different families and priors

# For family "gaussian" and iid normal prior
y_norm <- rnorm(n, mean = lin_pred, sd = 1)
dat_norm <- data.frame(Y = y_norm, X1 = x1, X2 = x2)

w05_mcmcglms <- mcmcglm_across_tuningparams(
   seq(from = 0.5, by = 0.5, length.out = 9),
   tuning_parameter_name = "w",
   formula = Y ~ .,
   family = "gaussian",
   data = dat_norm,
   n_samples = 10,
   burnin = 0
)