Skip to content

Latest commit

 

History

History
72 lines (62 loc) · 3.94 KB

NEWS.md

File metadata and controls

72 lines (62 loc) · 3.94 KB
Raw

Optim master release notes

  • The convergence tolerance computations have changed slightly

Optim v0.11.0 release notes

  • Optional scaling for inverse Hessian in L-BFGS
  • Support for initial step length guesses via LineSearches

Optim v0.10.0 release notes

  • Support for optimization on Riemannian manifolds
  • Support for optimization of functions of complex variables
  • New experimental KrylovTrustRegion method useful when cheap Hessian-vector products are available
  • Improved support for BigFloats
  • Add doc strings to methods
  • Drop support for Julia versions less than v0.6.0-pre

Optim v0.9.0 release notes

  • Fminbox: If an initial guess is on the boundary of the box, the guess is moved inside the box and a warning is produced, as opposed to crashing with an error.
  • Significant changes to the Non-, Once-, and TwiceDifferentiable setup; these now hold temporaries relevant to the evaluation of objectives, gradients, and Hessians. They also hold f-, g-, and h_calls counters
  • Refactor tests
  • Drop v0.4 support
  • Add limits to f-, g-, and h_calls
  • Improve trace for univariate optimization
  • Changed order of storage arrays and evaluation point arrays in gradient and Hessian calls
  • Skip v0.8.0 to allow fixes on Julia v0.5.0

Optim v0.7.6 release notes

  • Fix deprecations for *Function constructors
  • Fix depwarns on Julia master (v0.6)
  • Update references to new JuliaNLSolvers home for Optim+family

Optim v0.7.5 release notes

  • Various bug fixes
  • Deprecate DifferentiableFunction, TwiceDifferentiable in favor of OnceDifferentiable, TwiceDifferentiable
  • widen some type annotations (e.g. allow for multidimensional arrays as inputs again)
  • introduce allow_f_increases keyword in Optim.Options to allow objective to increase between iterations
  • New option in Optim.Options: allow_f_increases. Defaults to false, but if set to true, the solver will not stop even if a step leads to an increase in the objective.
  • Newton and BFGS: set initial step length to one. See 328.

Optim v0.7.3 release notes

  • OptimizationOptions is now unexported, and has been renamed to Options. Must be accessed as Optim.Options as a result.
  • Bug fixes to Nelder-Mead tracing.
  • Keywords with ! in them have been deprecated to version without. For example, linesearch! -> linesearch.
  • Failures in a line search now terminates the optimization with a warning and status of non-convergence. The results can still be accessed, but minimizer(res) will not represent a local minimum. See 275.

Optim v0.7.0 release notes

  • Refactor code internally to clean up code and allow more flexible use Optim
  • Switch to new (v.0.3) ForwardDiff
  • Make minimizer/minimum transition final
  • Fix dispatch bug for univariate optimization.
  • The line search functionality has been separated into a new package LineSearches.jl, see 277.
  • Make move to minimizer (the argmin) and minimum (objective at argmin) final: field names are now in sync with function based API.

Optim v0.6.1 release notes

  • Assess convergence in g before iterating to avoid line search errors if initial_x is a stationary point
  • Fix trace bug in LevenbergMarquardt.
  • Add ForwardDiff AD functionality to NewtonTrustRegion
  • Make documentation even more noticable in README.md

Optim v0.6.0 release notes

  • Various bug fixes
  • Added box constraints to Levenberg Marquardt
  • Changed old solver: Nelder-Mead w/ Adaptive parameters
  • Julia v0.5 deprecation fixes
  • Added NewtonTrustRegion solver 238 245
  • Added ParticleSwarm solver 218
  • Added documentation generated by Documenter.jl, see PR225.
  • Added NEWS.md