FactorHet_init {FactorHet}R Documentation

Arguments for initializing FactorHet

Description

A set of arguments that govern the initialization of FactorHet. Use FactorHet_control to set arguments around estimation. FactorHet_mbo ignores many of these arguments as it uses a single fixed initialization set by FactorHet_mbo_control. All arguments have default values.

Usage

FactorHet_init(
  short_EM = FALSE,
  short_EM_it = 40,
  short_EM_init = "random_member",
  short_EM_pi = NULL,
  force_rep = FALSE,
  verbose = FALSE,
  short_EM_beta_method = "cpp",
  short_EM_cg_it = 10,
  nrep = 5,
  debug_repeat = FALSE,
  plot_repeat = FALSE,
  return_all = FALSE
)

Arguments

short_EM

A logical value indicating whether "short EM" should be used. The default value is FALSE. TRUE indicates a "short EM" should be followed. That is, run multiple short runs of EM with random initializations and then proceed with the best for full initialization. Biernacki et al. (2003) provides more discussion. If FactorHet_control has init_method = "short_EM", this will override this setting.

short_EM_it

A numerical value of the number of iterations to use for each "short" run of the EM algorithm. The default is 40.

short_EM_init

An argument that sets the initialization procedure for "short EM". It must be some non-deterministic procedure that is valid in FactorHet_control. The default is "random_member".

short_EM_pi

An argument for the maximum number of iterations for the moderator updates to use for each "short" run of the EM algorithm. The default is NULL.

force_rep

A logical value for whether to repeat the algorithm if K=1. The default is FALSE and it should be used only for debugging.

verbose

A logical value to print more information about the progress of each iteration. The default is FALSE.

short_EM_beta_method

An argument for the update method for \beta to use for each "short" run of the EM algorithm. The default is "cpp".

short_EM_cg_it

An argument for the number of conjugate gradient iterations to use if short_EM_beta_method = "cg".

nrep

An integer value of the number of random iterations or runs of "short EM" should be used. The default value is 5.

debug_repeat

A logical value for whether to debug the repeated runs. The default is FALSE.

plot_repeat

A logical value for whether to plot the trajectory of the log-posterior for each run. The default is FALSE.

return_all

A logical value for whether to return all repetitions of the model versus the one with the highest log-posterior. The default is FALSE.

Value

FactorHet_init returns a named list containing the elements listed in "Arguments".

References

Biernacki, Christophe, Gilles Celeux, and Gérard Govaert. "Choosing Starting Values for the EM algorithm for Getting the Highest Likelihood in Multivariate Gaussian Mixture Models." 2003. Computational Statistics & Data Analysis. 41(3-4):561-575.

Examples

str(FactorHet_init())


[Package FactorHet version 1.0.0 Index]