optimize_latentfield_trustoptim: Optimize the conditional posterior of the latent Gaussian...

In awstringer1/casecrossover: Approximate Bayesian Inference for Case Crossover Models

Description

Compute the conditional mode W-hat for one value of the hyperparameters theta. Uses trust region methods implemented in trustOptim::trust.optim. Makes use of the sparsity of the Hessian.

This is a wrapper for optimize_latentfield_trustoptim() that computes the conditional mode in parallel over a user-specified grid of theta values.

Parallelization only works on linux and mac; the "parallel" package needs to work. If you're on windows, you'll just have to wait a little longer (long enough to go to the store and buy a serious computer maybe?).

Usage

optimize_latentfield_trustoptim(
  theta,
  model_data,
  hessian_structure = NULL,
  Q = NULL,
  optcontrol = NULL
)

optimize_all_thetas_parallel(
  thetagrid,
  model_data,
  hessian_structure = NULL,
  optcontrol = NULL,
  doparallel = TRUE
)

Arguments

theta	Vector, representing one configuration of hyperparameters (optimize_latentfield_trustoptim). List of vectors, each representing one configuration of hyperparameters (optimize_all_thetas_parallel).
model_data	ccmodeldata object as output by model_setup().
hessian_structure	Optional, provide the sparsity structure of the Hessian. This doesn't change with different thetas, so when doing all hyperparameter configurations, only need to compute this once. It will be computed if not provided.
Q	Optional, provide pre-computed Q matrix for this theta. Will be computed if not provided.
optcontrol	Optional. Override the casecrossover default control parameters for trust.optim(). See cc_control()$opt_control.
thetagrid	A NIGrid object which contains the grid of theta values to be optimized for.
doparallel	Logical. Use parallel::mclapply (TRUE) or lapply (FALSE) for execution? Former executes in parallel, latter in series. Default TRUE.

awstringer1/casecrossover documentation built on March 11, 2021, 4:41 a.m.