commit da6bc1c0ba3ad8d98b766a2cff78fc9d5d027b91
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2011-01-29 01:24:29 +0100

    Fix R CMD CHECK warnings.

commit 735ba47c9ff0b32c68457db143aa0831fb611583
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2011-01-29 01:21:42 +0100

    New release.

commit ccc9d1f3cc9bdb5cf724bd31a00d2bdee5fc90fd
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-12-20 16:32:42 +0100

    * Add documentation for 'diag.value' control option.
    
    * Small efficiency updates to cma_es(). Most notable change is that
      the generation of new individuals is now completely vectorized.
    
    * Better handling of box constraints. Penalized values are no longer
      eligible to be the 'current best'.
    
    * Add a new function extract_population() which can reconstruct the
      population at any iteration if the diagnostic options are set.

commit a2de7f3441fba004b608f858883d92ecc9bba1fc
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-09-10 15:22:09 +0200

    Allow vectorized fitness functions
    
    If the fitness function can accept multiple parameter settings at once,
    we can now pass a complete population in. Using this option it is
    trivial to parallelize the fitenss evaluation. This is especially
    important when optimizing expensive functions.

commit ba0222f98f33a35cca440b69da3ac109494892a4
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-14 00:43:18 +0200

    Fix bug when maximizing.
    
    When maximizing, the best fitness value was not chosen based on
    fn()*fnscale.
    
    Additionally some unit tests were refined.

commit 933f5c7ac9315fa4777fb8379d68da83a3697385
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-13 22:39:49 +0200

    More name fixes to be compatible with optim()
    
    USER VISIBLE CHANGE: When fn() is called, x will be a named vector if
      par has names.
    
    RATIONAL: optim() behaves the same way and we want to be a drop in
      replacement for optim() wherever possible.
    
    WHAT TO CHANGE: Nothing if you do not rely on being able to access
      elements of x by name in fn(). If you have relied on this and have
      manually set the names in fn(), you can now remove that piece of
      code and only set the names on par once before calling cma_es().

commit 2e98dfecb4b3d6e50771dbe6bd038d4c07077b85
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-13 22:07:40 +0200

    Always track best solution based on actual parameters used.
    
    Instead of penalizing infeasable parmater combinations and then
    caluclating the best solution based on this, use the actual function
    values to track best solution.
    
    USER VISIBLE CHANGE: par element of result list should always be
      feasible, even if the optimum is located on the border of the feasible
      region. The values displayed by the trace message may differ from those
      returned, because the continue to use the penalized values.
    
    RATIONAL: Previous behaviour was the result of a bad design choice and
      is now considered a bug. While certain solutions may 'virtually' lie
      outside the feasible region, all parameter settings returned by
      cma_es() should be feasible.
    
    WHAT TO CHANGE: Nothing.

commit 58360087a6255a784a67cd1153c79810ebbb2af1
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-13 21:46:00 +0200

    Changes 'flat fitness function' warning to a message.
    
    Instead of throwing a warning every time we find a flat piece of fitness
    landscape, we only show a message if trace=TRUE in the control list.
    
    USER VISIBLE CHANGE: cmaes() will no longer warn if the fitness
      landscape is flat and sigma is increased.
    
    RATIONAL: These warnings are usualy harmless and only confused the
      casual user.
    
    WHAT TO CHANGE: Nothing unless you want to continue to see the message.
      If so, please set trace=TRUE in the control list.

commit 8cf5c28ee4c9da3fb095463db910575afac0dca6
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-13 19:21:14 +0200

    Remove unnecessary preallocation of internal matrix.

commit 78987a16f22f5f5c02622c2ab0df6bc85d7f1803
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-13 17:20:32 +0200

    Bump version to 1.0-9

commit d7a6ba07af46b7375c50c6d66cac79be7fc1464d
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-07-13 17:18:45 +0200

    Copy names of par to final (optimal) parameter vector
    
    optim() copies the names of the par vector to the result parameter
    vector, so we will do the same.

commit 1113809ce01e16a9473899832d34193f603419e9
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-06-11 23:15:35 +0200

    Bump version number.

commit 7a56e13960be57bfda3073ef3e0832450d0231d2
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-06-11 23:00:49 +0200

    Fix bug in stopfitness interepretation.
    
    stopfitness was previously compared to value * fnscale. Now it is
    (correctly) interpreted with respect to the value. The actual stop
    condition is:
    
      value * fnscale <= stopfitness * fnscale
    
    This means that for minimization cma_es will stop whenever the current
    best value is equal to or below the given stopfitness. If maximizing
    then the best function value must be greater than or equal to the given
    stopfitness.

commit 49e3a20cdc1f43bc9c0e065a9d2e59b902b19fec
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-05-23 14:56:07 +0200

    Release of 1.0-5

commit a38b9c24f23bc0f2940e9ccc06c053f9ba2e4553
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-05-23 14:55:46 +0200

    Fix incorrect case in filename of runner.r.

commit 3ca5e511efbba866032f1cfba19c484621d3b2c7
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-05-23 14:53:52 +0200

    Release of 1.0-4

commit d37473868499a366b317c4584cb5b919c2651a92
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-05-23 14:51:17 +0200

    Update unit test to use cma_es() instead of deprecated cmaES().

commit 38086c79cc86bcab4cda460aafb31b240a93546c
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-05-23 14:50:48 +0200

    Cleanup minimal set of test functions.

commit 8b2cc34c8d7c80f6f56fba97c8518d139814c2f3
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-05-23 14:50:13 +0200

    Fix constraint penalty handling.

commit 7d5b45fd5cd645b473b892822961ff19cd8c0a92
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-03-08 13:10:39 +0100

    Small changes to examples.

commit 1b7c78937211ce41a1557931e8051b2535cd9867
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2010-03-08 12:29:02 +0100

    Fix constraint handling.

commit 00f573e7e7abc5542897dff9808ed95d5609f72d
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2009-11-13 00:51:05 +0100

    More Roxygen cleanup.
    
    Cleanup roxygen comments and remove some of the test function cruft.
    Also added bump-version script and fixed Makefile.

commit 013ce310ddfe11b9f4628166902d421b0f5ab8d4
Author: Olaf Mersmann <olafm@statistik.tu-dortmund.de>
Date:   2009-11-10 19:51:49 +0100

    Switch to Roxygen.
