hier_models: MCMC functions for the hierarchical versions of Normal Theory...
In jrlewi/brlm: Bayesian Restricted Likelihood Methods

Description Usage Arguments Details

MCMC functions for the hierarchical versions of Normal Theory Model, corresponding restricted version, and heavy-tailed version used in the paper.

Function to fit restricted likelihood version of hierarchical model in paper

fn.hier.one.rep(
  y,
  X,
  XtX,
  v1,
  v2,
  bstar,
  Beta,
  betalMat,
  Z,
  mu_rho,
  psi_rho,
  rho,
  step_logbstar,
  mu_rho_step,
  psi_rho_step,
  rho_step,
  step_Z,
  Sigma0Inv,
  nGroups,
  p,
  abc = FALSE
)

hierNormTheoryLm(
  y,
  X,
  nkeep = 10000,
  nburn = 1000,
  mu0,
  Sigma0,
  a0,
  b0,
  mu_bstr,
  psi_bstr,
  swSq = 1,
  w1,
  w2,
  a_psir,
  b_psir,
  step_logbstar,
  mu_rho_step,
  psi_rho_step,
  rho_step,
  step_Z
)

hierNormTheoryRestLm(
  y,
  X,
  regEst = "Huber",
  scaleEst = "Huber",
  nkeep = 10000,
  nburn = 1000,
  mu0,
  Sigma0,
  a0,
  b0,
  mu_bstr,
  psi_bstr,
  swSq = 1,
  w1,
  w2,
  a_psir,
  b_psir,
  maxit = 400,
  step_logbstar,
  mu_rho_step,
  psi_rho_step,
  rho_step,
  step_Z,
  abc = FALSE,
  bandwidth = NULL,
  iter_check = 1000,
  min_accept_rate = 0.1,
  bw_mult = 1.2
)

fn.compute.marginals.hierModelNormal(
  betalsamples,
  sigma2lsamples,
  yhold,
  Xhold
)

tdensity(y, mean, sigma, nu)

fn.compute.marginals.hierModelTmodel(
  betalsamples,
  sigma2lsamples,
  yhold,
  Xhold
)

fn.one.rep.tHierModel(
  y,
  X,
  v1,
  v2,
  bstar,
  Beta,
  betalMat,
  vlList,
  Z,
  mu_rho,
  psi_rho,
  rho,
  step_logbstar,
  mu_rho_step,
  psi_rho_step,
  rho_step,
  step_Z,
  Sigma0Inv
)

hier_TLm(
  y,
  X,
  nkeep = 10000,
  nburn = 1000,
  mu0,
  Sigma0,
  a0,
  b0,
  mu_bstr,
  psi_bstr,
  swSq = 1,
  w1,
  w2,
  a_psir,
  b_psir,
  nu,
  step_logbstar,
  mu_rho_step,
  psi_rho_step,
  rho_step,
  step_Z
)

`y, X`	lists of group level responses and design matrices
`XtX`	list of X'X - for all the groups (input for efficiency)
`v1, v2`	parameters for beta on b_start. mu_bstr is the mean and psi_bstr (originally had a prior) is the precision of the beta prior for b^*
`bstar`	desc
`Beta`	desc
`betalMat`	desc
`Z`	desc
`mu_rho`	desc
`psi_rho`	desc
`rho`	desc
`step_logbstar, mu_rho_step, psi_rho_step, rho_step, step_Z`	tunning parameters for MH Steps
`abc`	new option, defaults to FALSE, if TRUE then an Approximate Bayesian Computation method version is fit
`nkeep, nburn`	number of MCMC iterations to keep, number for burn in.
`Sigma0`	is the 'variance' matrix of beta b_i~N(mu0,b^*Sigma0)
`a0, b0`	prior parameters for sigma2
`swSq`	default to 1.
`w1, w2, a_psir, b_psir`	parameters definining prior for rho. In detail: rho~beta(mean=mu_rho, precision=psi_rho), mu_rho~beta(w1,w2) and psi_rho~gamma(a_psir, b_psir)
`regEst`	Regression estimator on which to condition . Either Huber or Tukey.
`scaleEst`	Scale estimator on which to condition('Huber' is only option here)
`bandwidth`	for the abc kernel, scalar or vector of length(X) specifying the abc bandwidth for each group.
`betalsamples`	the array of betals: in the specific format: the 3rd dimension is the groups. columns represent samples, row represent slopes
`yhold`	list of holdout samples
`Xhold`	list of design matrices
`mean`	center of t distribution
`sigma`	scale of t distribution
`nu`	fixed degrees of freedom for assumed t-distribution
`sigma2Int`	is the vector of sigma2i initial values
`mu0:`	prior mean of each beta
`y`	data
`sigma2lsamples:`	columns represnt groups, rows represent samples

fn.hier.one.rep is one rep for the full normal thoery model. heirNormTheoryLm uses fn.hier.one.rep for the complete MCMC of full normal thoert model. hierNormTheoryRestLm does the same for the restricted versions. The corresponding t-model versions are handedl by fn.one.rep.tHierModel and hier_TLm.

Details of the model are describted in the paper.

for abc version method - see ()

jrlewi/brlm documentation built on March 17, 2021, 1:10 a.m.