Is JOONE still supported?

2009-02-09
2013-04-16
  • Tom Johnson
    Tom Johnson
    2009-02-09

    I am working on a neural network for a thesis for my degree.  I wanted to use an open source framework.  I am looking at JOONE, and also Encog ( http://www.encog.org ). 

    I started with JOONE, but there seem to be many bugs.  And the threading is really frustrating.  It seems like the threading is actually making JOONE run slower on a multiprocessor machine.  Which does not make sense to me. 

    It seems like there's not been an update for 2 years, is JOONE still an active project?

     
    • UeliHofstetter
      UeliHofstetter
      2009-02-09

      hi there,

      the project is kind of desolated currently. however it offers a very comprehensive set of (unfortunately undocumented)features, you will probably not find in other projects (i.e. afaik the encog project is the sample code for a really introductory book on neural networks using java).

      so if you really want some advanced features and if you aren't scared of hunting one or the other bug, i recommend you to dive into joone (get the latest version from cvs). otherwise you're probably better of using another framework.

      you may use your thesis to polish up joone (involvement is currently extremly welcomed ;-))?

      kind regards

      u

       
    • Tom Johnson
      Tom Johnson
      2009-02-10

      I actually read up on Encog a bit.  You are correct, its based off of the source code from a Neural Network book.  That is where it started, but its grown since then.  The author basically organized all the changes people were requesting/making to the book's code and created the Encog project. 

      Still evaluating JOONE is well.  This is really just my opinion.  But JOONE seems terribly over engineered.  Which makes it a bit slow.  Spawning a separate thread for each layer of the neural network just seems excessive.  Plus, thats a ton of threads, those need to be pooled, even if you were going for such a thread-centric strategy.  Also why use threads on running the neural network?  Usually thats pretty fast.  Its training that you need some sort of a threaded method.  Not that backprop can be threaded all that well.

      Also the Joone GUI tool not running on Vista is really annoying, I have to use a VMWare XP image just to use it.  I thought it was just broke in general.  But several other users here pointed out it just does not work with XP.  Any idea when that will be upgraded?  I am not much of a GUI programmer myself, so I have no idea what that takes.

      My suggestion for JOONE would be to pretty much gut the threading model.  I can't see what threads even add to JOONE, they slow it down.  

      Is there any sort of a road map for where future enhancements of JOONE are going?  Is performance/threading a priority?