NUTS: No-U-Turn sampler

In deepfriar/NUTS-mirror: No-U-Turn sampler for HMC

Description

No-U-Turn sampler

Usage

NUTS(
  theta,
  f,
  grad_f,
  n_iter,
  M = diag(length(theta)),
  M_adapt = 50,
  M_chol = Matrix::chol(M),
  M_inv = Matrix::solve(M),
  delta = 0.5,
  max_treedepth = 10,
  eps = 1,
  find = TRUE,
  verbose = TRUE
)

Arguments

theta	Initial value for the parameters
f	log-likelihood function (up to a constant)
grad_f	the gradient of the log-likelihood function
n_iter	Number of MCMC iterations
M	the HMC mass matrix. Defaults to diag(length(theta)).
M_adapt	Parameter M_adapt in algorithm 6 in the NUTS paper
M_chol	The cholesky decomposition of M. Default Matrix::chol(M).
M_inv	The inverse of M. Default Matrix::solve(M).
delta	Target acceptance ratio, defaults to 0.5
max_treedepth	Maximum depth of the binary trees constructed by NUTS
eps	Starting guess for epsilon
find	Allow NUTS to make a better guess at epsilon? Default TRUE.
verbose	logical. Message diagnostic info each iteration? Default TRUE.

Value

Matrix with the trace of sampled parameters. Each mcmc iteration in rows and parameters in columns.

deepfriar/NUTS-mirror documentation built on Dec. 19, 2021, 10:08 p.m.