[mlpack] mlpack 2.1.0 released

Ryan Curtin ryan at ratml.org
Mon Oct 31 11:57:57 EDT 2016


Hello there,

Today I tagged and uploaded mlpack 2.1.0 as a new release.  There are
some new method additions, mostly from the great work done as a part of
GSoC 2016, and a few minor reverse-incompatible API changes, mostly in
the long-form command line options to programs.  Key changes:

 * Add a multitude of new tree types from Mikhail Lozhnikov's GSoC
     project.
 * Add mlpack_preprocess_* programs for preprocessing data, from Keon
     Kim's GSoC project.
 * Add approximate nearest and furthest neighbor extensions from Marcos
     Pividori's GSoC project.
 * Add LSH multiprobe support from Yannis Mentekidis' GSoC project.
 * Add new algorithms for approximate furthest neighbor search.
 * Add gradient descent optimizer.
 * Various bugfixes.

You can find the release at

  http://www.mlpack.org/files/mlpack-2.1.0.tar.gz

and a more detailed list of changes at

  http://www.mlpack.org/history.html

and if you really like changes, you can see every commit since 2.0.3
here:

  https://github.com/mlpack/mlpack/compare/mlpack-2.0.3...mlpack-2.1.0

Next, a quick look at what's in the future for mlpack 3.0.0 (this is the
version it will have to be because it will have reverse
incompatibilities):

 * Better OpenMP support for various implementations (so, speedups).
 * Better usage of vectorization in mlpack code (again, more speedups).
 * Auto-tuning LSH from Yannis's GSoC project, which will have a nice
     interface.  Maybe some other auto-tuning too!
 * Bindings to other languages, such as Python, MATLAB, and R
     (finally!).

Those are just my priorities, so it's possible others have other
priorities not listed here that will also be incorporated.  :)

In addition, I recently heard that a group from DTU has implemented a
lot of code on top of the mlpack LSH implementation, so I am hoping to
get in touch with them to get them to contribute it upstream.  I think
that would be a really nice addition to mlpack's already complete
support for nearest-neighbor-searching tasks.

If you have any problems or would like to contribute your own code, feel
free to get in touch on Github at https://github.com/mlpack/mlpack or on
IRC in #mlpack on freenode or even just send an email to this mailing
list.

-- 
Ryan Curtin    | "We need some time for some things to happen!"
ryan at ratml.org |   - Bells


More information about the mlpack mailing list