Skip to contents

Create line search configuration

Source: R/config.R
mle_config_linesearch.Rd

Extends gradient configuration with backtracking line search parameters. Line search adaptively finds step sizes that improve the objective.

Usage

mle_config_linesearch(
  max_step = 1,
  backtrack_ratio = 0.5,
  max_iter_ls = 10L,
  min_step = 1e-08,
  norm = function(x) max(abs(x)),
  max_iter = 100L,
  abs_tol = NULL,
  rel_tol = 1e-05,
  trace = FALSE,
  debug = FALSE,
  debug_freq = 1L
)

Arguments

max_step

Maximum step size per iteration (numeric, default: 1.0)

backtrack_ratio

Backtracking multiplier, must be in (0,1) (default: 0.5)

max_iter_ls

Maximum line search iterations (integer, default: 10)

min_step

Minimum step size threshold (numeric, default: 1e-8)

norm

Distance measure function (default: max absolute value)

max_iter

Maximum iterations (integer, default: 100)

abs_tol

Absolute tolerance (numeric or NULL, default: NULL to use rel_tol)

rel_tol

Relative tolerance (numeric, default: 1e-5)

trace

Store optimization path (logical, default: FALSE)

debug

Print debug information (logical, default: FALSE)

debug_freq

Debug output frequency (integer, default: 1)

Value

An mle_config_linesearch object

Examples

# Conservative line search
config <- mle_config_linesearch(
  max_step = 0.5,
  backtrack_ratio = 0.8
)

# Aggressive line search
config <- mle_config_linesearch(
  max_step = 10.0,
  backtrack_ratio = 0.3,
  max_iter_ls = 20
)