v1.4.1

Optim v1.4.1

Diff since v1.4.0

Merged pull requests:

• d -> obj in bfgs with initial_stepnorm !== nothing (#935) (@chriselrod)

v1.4.0

Optim v1.4.0

Diff since v1.3.0

Closed issues:

• Ensure alphamax linesearch is set for AcceleratedGradientDescent (#458)
• how to make optim work with Measuements (#823)
• Preconditioner example is not working (#837)
• Add support for OffsetArrays (#868)
• Optim.ldiv! is only defined on arrays (#870)
• Optim.trace() fails with MethodError: no method matching length(::Nothing) (#895)
• Importing Optim.jl causes really unexpected behaviour in unrelated code (#900)
• NaN given in parameter vector (#921)
• LBFGS: Does Optim.jl also implement an improvement over the original routine (which scipy.optimize does)? (#923)
• How to use Optim.jl's Optim.OnceDifferentiable in the context of LBFGSB.jl ? (#926)
• How do you do LBFGS() within a bound box for 2 dimensions? (#930)

Merged pull requests:

• relax type constraints to support Measurements (#913) (@longemen3000)
• Stop being a ☠️ (fix #900, ldiv! and dot pirates Base). (#916) (@pkofod)
• Fix eachindexing and dots for NelderMead. (#917) (@pkofod)
• NLOpt -> NLopt (#928) (@blegat)
• BFGS GPU Support (#931) (@avik-pal)
• CompatHelper: bump compat for "FillArrays" to "0.12" (#932) (@github-actions[bot])

v1.3.0

Optim v1.3.0

Diff since v1.2.4

Merged pull requests:

• Allow the choice between fg or f when evaluating the trust region sub… (#911) (@pkofod)

v1.2.4

Optim v1.2.4

Diff since v1.2.3

Closed issues:

• Optim.minimizer returns the initial starting point (#901)
• How to set lower and upper bound for Newton's algorithm? (#905)

Merged pull requests:

• Fix LBFGS when rho is infinite and fix initial convergence change (#908) (@pkofod)

v1.2.3

Optim v1.2.3

Diff since v1.2.2

Merged pull requests:

• Change order of ldiv and - in sparse Newton. (#898) (@mikkoku)
• Fix some v1.6 stuff (#899) (@pkofod)

v1.2.2

Optim v1.2.2

Diff since v1.2.1

Closed issues:

• TwiceDifferentiable and FMinbox seem incompatible (#600)
• Wrongly reporting success (#873)
• Fminbox problem with integers (#885)

Merged pull requests:

• CompatHelper: bump compat for "FillArrays" to "0.11" (#897) (@github-actions[bot])

v1.2.1

Optim v1.2.1

Diff since v1.2.0

Closed issues:

• Optim 1.0, 1.1 and Julia 1.0 (#863)
• BarrierWrapper has no field f_calls (#865)
• show_trace panic with Optim.Brent() (#867)
• Documentation: Something is wrong with the Matlab Simplexer example: (#871)
• optimize passes NaN as x value to function (#874)
• Does L-BFGS support mini-batch gradient decent? (#875)
• L1 regularization with box constraint in Optim ? (#877)
• particle swarm candidate solution (#880)
• show_trace fails with Brent (#881)

Merged pull requests:

• Small typo in doc, code doesnt run without .+ (#851) (@eirikbrandsaas)
• Don't repeatedly calculate first element of the swarm in ParticleSwar… (#869) (@pkofod)
• Test that tracing works for univariate optimization. (#879) (@pkofod)
• Create CompatHelper.yml (#886) (@briochemc)
• Add Github Action CI (#888) (@briochemc)
• make some utils gpu friendly (#889) (@piever)
• CompatHelper: bump compat for "FillArrays" to "0.10" (#890) (@github-actions[bot])
• Fix reset of LBFGS. (#891) (@pkofod)

v1.2.0

Optim v1.2.0

Diff since v1.1.0

Closed issues:

• Fminbox reset! bug: d not defined (#864)
• Error in using Optim with Julia v1.0: ERROR: LoadError: LoadError: UndefVarError: InPlaceObjectiveFGHv not defined (#866)

v1.1.0

Optim v1.1.0

Diff since v1.0.0

Closed issues:

• Doc example with complex numbers not working (#780)
• BFGS reset! typo (#859)
• Fminbox evaluates outside the box (#861)

Merged pull requests:

• Make it easier to use IPNewton if you only have bounds. (#858) (@pkofod)
• Fix out-of-box in Fminbox. (#862) (@pkofod)

v1.0.0

Optim v1.0.0

Diff since v0.22.0

Closed issues:

• usage with JuMP? (#107)
• Generalize Optim.jl to be an interface for nonlinear optimization (#309)
• Multidimensional arrays (#399)
• ProgressMeter (#442)
• Missing docstrings (#469)
• L-BFGS-(B) (#521)
• Finite differencing should respect box constraints (#541)
• Todo: SAMIN (#555)
• Some questions about convergence assessment and results (#574)
• Going below sqrt(eps) (#631)
• Project manifold tangent when we reset search direction (#649)
• Rename optimize to minimize? (#685)
• Preallocate caches in Fminbox to avoid unnecessary value_gradient!! calls (#704)
• No default objective type for IPNewton (#711)
• Fminbox() using parameters outside of the box specified (#712)
• only_fgh! doesn't work if algorithm does not need Hessian (#718)
• Bug in Stiefel_CholQR (#752)
• Intermittent assertion error in HagerZhang (#802)
• Success when none of convergence measures is satisfied (#806)
• feature request: default to gradient-based method when user passes fg! (#816)
• Use AbstractConstrainedOptimizer as the parent type for all the constrained optimizers? (#818)
• Type instability (#820)
• Confusing behavior of f_abstol with Fminbox (#821)
• Univariate optimization does not stop when callback returns true (#822)
• Optim: No method matching iterate (#824)
• Brent still broken for flat functions (#827)
• Comment doesn't really seem to reflect logic in code (#830)
• Computing the inverse of Hessian in Newton's method (#832)
• Fminbox optimization does not exit when callback returns true (#834)
• Particle Swarm algorithm fails in very simple objective function (#835)
• ParticleSwarm does not respect Optim.Options(iterations=x) (#836)
• Missing website (#841)
• ERROR: ArgumentError: Value and slope at step length = 0 must be finite. (#842)
• Problems with BigFloat (#844)
• indexing issue with ArrayParition arguments (#848)
• NealderMead recalculates same points: (#852)
• Functor Support with Univariate analysis (#853)
• Warning (again) on LinearAlgebra.dot (#855)

Merged pull requests:

• WIP: Corrected Stiefel_CholQR (#753) (@jagot)
• Add callback stops for univariate optimization. (#831) (@pkofod)
• Test fgh! (#838) (@pkofod)
• Update .travis.yml (#839) (@pkofod)
• Bump FillArrays version (#840) (@JeffFessler)
• Make promote_obj noop for ipnewton and twicediffed (#845) (@pkofod)
• Add methods for (Not)InplaceObjective to optimize when not method is … (#846) (@pkofod)
• Allow that f can increase from iteration to iteration (#847) (@pkofod)
• Fix project_tangent! and retract! args in manual (#849) (@thisrod)
• Don't re-create the cache in every iteration of Fminbox as it calls the objective too often (#850) (@pkofod)
• Don't use Function signatures. (#854) (@pkofod)
• Fix some outstanding Fminbox issues (#856) (@pkofod)