Skip to contents

Changelog

Source: NEWS.md

mize 0.3.0

This release improves optimizer robustness, input validation, and progress diagnostics.

New features

  • New function: check_mize_gradient(), which compares an analytic gradient with a finite-difference approximation.
  • mize() now returns status fields: converged, status, and message, plus explicit best/last result fields (best_par, best_f, last_par, and last_f).
  • mize(), mize_step(), mize_step_summary(), and stored progress now report accepted Hessian and inverse-Hessian callback counts as nh and nhi.
  • With store_progress = TRUE, mize() now exposes optional line-search reason, selected-point provenance, local callback counts, initial scale, and exact-Newton direction provenance.

Bug fixes and minor improvements

  • Backtracking line searches no longer report line_search_failed prematurely when small steps are needed to reach the requested convergence tolerance.
  • Line searches now reuse step0 after returning a zero step when step_next_init depends on previous search results. This allows momentum methods to recover on the next iteration.
  • Bold Driver now honors step0 instead of always starting at 1. It requires a positive finite numeric scalar, with NULL retaining the default of 1.
  • Bold Driver now accepts step sizes below its former minimum, avoiding premature line-search failures when smaller steps are needed. After a zero or rejected gradient step, the next search starts with a positive step size so gradient descent can resume after a momentum update.
  • BFGS and SR1 now accept one-dimensional inverse-Hessian vectors consistently with equivalent one-by-one matrices.
  • The cg_update = "HZ+" safeguard now uses Euclidean norms as specified by its formula. This correction can change optimization trajectories and iteration counts.
  • check_mize_gradient() rejects unrepresentable or non-finite coordinate perturbations and dimensioned callback results instead of reporting misleading agreement or flattening matrices.
  • Gradient norms and step lengths avoid intermediate overflow and underflow when the Euclidean norm is representable. CG falls back to steepest descent when its direction update produces unusable arithmetic or fails to give a descent direction.
  • store_progress = TRUE has lower time and allocation costs on long runs.
  • Classical Momentum with numeric mom_schedule = 0, and NAG with the "nsconvex" schedule, nest_q = 1, and nest_convex_approx = FALSE, no longer make redundant function and gradient calls. In these configurations, restart is ignored and progress no longer includes mu.
  • DBD now returns structured fn_inf or gr_inf failures when a later objective or gradient is non-finite, and par_inf if an update would produce non-finite parameters. Such an update is rolled back before callbacks can observe it. DBD also validates its initial step controls and safeguards named initial-step estimates using the current parameter scale.
  • Momentum schedules now reject malformed or non-finite configured values and function results. The "ramp" and "switch" schedules require numeric mom_init and mom_final values, and adaptive restart safely rejects non-finite comparisons instead of raising a control-flow error.
  • Stateful optimization now handles lifecycle dependencies consistently, and global function and gradient evaluation limits are enforced across optimizer and line-search callbacks, including truncated Newton inner iterations.
  • Optimizer inputs and objective, gradient, Hessian, and inverse-Hessian callback results now receive consistent early validation and clearer errors.
  • Quasi-Newton updates and exact-Newton directions now use safer fallbacks when curvature information or Hessian factorization is unsuitable.
  • NEWTON and L-BFGS now preserve plain vector parameters across multiple iterations when a supplied inverse-Hessian callback returns a full matrix.
  • Wolfe line searches now require an explicitly numeric step0 to be a positive finite scalar; string initializers are unchanged.
  • Line searches using weak Wolfe curvature now accept equality at the curvature boundary. The optional Hager-Zhang initializer probe now counts toward the line search’s local function and combined evaluation limits. Hager-Zhang initializer arithmetic also safely handles non-finite values and uses the specified Euclidean gradient norm.
  • Bracketed Wolfe line searches now avoid repeated objective and gradient callbacks when floating-point step changes cannot produce a new parameter vector.
  • A line search that selects no usable step and produces no complete optimizer transition now reports line_search_failed instead of tolerance convergence. Bold Driver now follows this contract, reports line-search diagnostics, and avoids callbacks at trial step lengths that reproduce an evaluated parameter vector.
  • Bold Driver’s alpha diagnostic reports the selected gradient step length, and is zero when no gradient step was selected. Its ls_max_fn = 0 limit is handled consistently whether progress is stored.
  • mom_type = "nesterov" now applies to momentum schedules attached to methods other than "Momentum", including DBD. These configurations previously used classical momentum silently.

mize 0.2.5

CRAN release: 2026-02-09

Bug fix release.

Bug fixes

  • Removed LazyData from the DESCRIPTION file.
  • Some flaky unit tests were removed.

mize 0.2.4

CRAN release: 2020-08-30

Bug fix release.

Bug fixes.

  • If using line_search = "backtracking" with a specified step_down parameter, an incorrectly large number of gradient calculations was being reported.
  • The documentation now specifies that if you don’t provide a step_down argument with line_search = "backtracking", interpolation using function and gradient evaluations is carried out. To get a typical Armijo-style backtracking line search, specify a value for step_down (e.g. step_down = 0.5 to halve the step size), and only function evaluations are used.

mize 0.2.3

CRAN release: 2019-12-05

A patch release to fix an incompatibility with R-devel.

Bug fixes

  • Fixed a bug where class was being checked directly and a scalar value was assumed. The correct behavior is to use methods::is.

mize 0.2.2

CRAN release: 2019-07-11

A patch release for a bug fix.

Bug fixes

  • Fixed a bug where if the maximum number of function evaluations for the Schmidt line search is exceeded (controlled by the ls_max_fn parameter), a 'bracket_step' not found error could result. Thank you to reporter Charles Driver.
  • Fixed a couple of vignette links that were missing the “http://” at the front.

mize 0.2.1

CRAN release: 2019-04-14

A patch release to fix an incompatibility with R-devel.

Bug fixes

  • Fixed an error with bold driver and back-tracking line search where the stage was being incorrectly checked.

mize 0.2.0

CRAN release: 2018-09-14

New features

  • New method: Truncated Newton (method = "TN"). Can be controlled using the tn_init and tn_exit options.
  • New method: SR1 (method = "SR1"), falling back to the BFGS direction if a descent direction is not found.
  • New option preconditioner, which applies to the conjugate gradient and truncated newton methods. The only value currently available is preconditioner = "L-BFGS" which uses L-BFGS to estimate the inverse Hessian for preconditioning. The number of updates to store for this preconditioner is controlled by the memory parameter, just as if you were using method = "L-BFGS".
  • BFGS, SR1, L-BFGS methods will now make use of a user-supplied inverse Hessian function if provided. In the input fg list, supply a function hi, that takes the par vector as input. The function can return a matrix (obviously not a great idea for memory use), or a vector, the latter of which is assumed to be the diagonal of the matrix.
  • ls_max_alpha (for line_search = "More-Thuente" only): sets maximum value of alpha that can be attained during line search.
  • ls_max_alpha_mult (for Wolfe-type line search only): sets maximum value that can be attained by the ratio of the initial guess for alpha for the current line search, to the final value of alpha of the previous line search. Used to stop line searches diverging due to very large initial guesses.
  • ls_safe_cubic (for line_search = "More-Thuente" only): if TRUE, use the safe-guarded cubic modification suggested by Xie and Schlick.
  • cg_update = "prfr", the “PR-FR” (Polak-Ribiere/Fletcher-Reeves) conjugate gradient update suggested by Gilbert and Nocedal.

Bug fixes

  • An error occurred when checking if a step size was finite during line search.
  • DBD method didn’t use momentum when asked to.
  • Fix incorrectly specified conjugate gradient descent methods: Hestenes-Stiefel (cg_update = "hs"), Conjugate Descent (cg_update = "cd"), Dai-Yuan (cg_update = "dy") and Liu-Storey (cg_update = "ls").

mize 0.1.1

CRAN release: 2017-07-14

Initial release.