mle_local_search

Performs a local search to find the MLE, assuming the MLE is an interior point of the support and that an initial guess `theta0` that is near the MLE is provided. Use a global search method like `sim_anneal` to find a good initial guess.

Usage

mle_local_search(dir, theta0, loglike = NULL, options = list())

Arguments

dir

function, promising direction function

theta0

numeric, initial guess

options

list, options for the local search, see function description.

Value

an `mle` object with additional attributes `iter` and `converged` and optionally `path` if `trace` is TRUE.

Functions

• mle_local_search(): options

Fields

sup

function, domain of support for log-likelihood

eta

numeric, learning rate, defaults to 1

max_iter

integer, maximum number of iterations, defaults to 1000

max_iter_ls

integer, maximum number of iterations for the line search, defaults to 1000

abs_tol

numeric, tolerance for convergence, defaults to NULL (use rel_tol instead)

rel_tol

numeric, relative tolerance for convergence, defaults to 1e-5

r

numeric, backtracking line search parameter, defaults to 0.5

proj

function, projection function to enforce domain of support

norm

function, distance measure for convergence checks, defaults to the to the infinity norm.

debug

logical, output debugging information if TRUE; default FALSE

trace

logical, if TRUE store the path of the search in the `path` attribute of the output; default FALSE

line_search

logical, if TRUE, perform a line search; default TRUE in this case, learning rate `eta` refers to the maximum step size that can be taken per iteration.

debug_freq

integer, frequency of debug output, defaults to 1