We are happy to announce that Optim v0.6 has been released.

Since v0.5, we've seen some great additions and changes to the package. Much
of this has happened with the help of new and old contributors, and 
therefore
this version tag should really be dedicated to them. We've had nine 
contributors
in this minor tag (@pkofod, @ranjanan, @ahwillia, @bjarthur, @Ace139, 
@rgiordan,
@abieler, @matthieugomez, and @amroamroamro). This brings us up to 36 
contributors
in total (I'm not counting you @gitter-badger!), and 15 this year alone. It 
is
great to see so many volunteering their time for this open source project.

In this version we have the following changes
* Added NEWS.md
* Added proper documentation using Documenter.jl
* Changed old solver: Nelder-Mead w/ Adaptive parameters (breaking)
* Julia v0.5 deprecation fixes
* Added solver: NewtonTrustRegion
* Added solver: ParticleSwarm
* Added box constraints to Levenberg Marquardt
* Changed Automatic Differentiation to ForwardDiff
* Various bug fixes

Worth noting here is that we have some changes to the Nelder-Mead API. It is
documented in the new documentation.

We also have two new solvers: the NewtonTrustRegion solver, and the 
ParticleSwarm
solver. A lot of focus and effort has gone into first order solvers, so 
it's interesting
to see both a new zeroth and second order solver in Optim.

Box contrains in Levenberg-Marquardt was added, and we switched to 
ForwardDiff for
automatic differentiation, such that you can now obtain Hessians for the 
two Newton
solvers.

There's plenty of work in the pipeline, so stay tuned for more Optim 
improvements
in the future. We are looking into code refactoring to easy use of the code 
in more
flexible ways, better benchmarking, better testing, and more.

As always, please report bugs or feature requests to the issue tracker
(https://github.com/JuliaOpt/Optim.jl/issues), and join the chat at gitter
(https://gitter.im/JuliaOpt/Optim.jl).

Optim.jl / JuliaOpt

Reply via email to