[ 
https://issues.apache.org/jira/browse/MATH-1552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17357953#comment-17357953
 ] 

Gilles Sadowski commented on MATH-1552:
---------------------------------------

As of commit df927f0b674d8c6cd5f535c2995777a4dd7be101 ("master" branch):
 * {{MultiDirectionalSimplex}} pass all tests.
 * {{NelderMeadSimplex}} has 8 tests disabled.

For {{MultiDirectionalSimplex}}, the improvement was due to the increase of the 
initial simplex size (from 0.1 to 1).

Finding the global optimum is very sensitive to the (randomized) starting 
point.  {{NelderMeadSimplex}} often finds a local optimum whose objective value 
is quite close to that of the global optimum.

> Simplex optimizers fail standard test functions
> -----------------------------------------------
>
>                 Key: MATH-1552
>                 URL: https://issues.apache.org/jira/browse/MATH-1552
>             Project: Commons Math
>          Issue Type: Test
>    Affects Versions: 3.6.1
>            Reporter: Gilles Sadowski
>            Priority: Major
>             Fix For: 4.X
>
>
> In [commit 
> 3d2b2107b58ffc316c849da43734506b7fe04254|https://gitbox.apache.org/repos/asf?p=commons-math.git;a=commit;h=3d2b2107b58ffc316c849da43734506b7fe04254],
>  12 problems for exercising optimizers (originally provided in the 
> {{BOBYQAOptimizer}} tests suite) were added to the unit test classes for 
> {{NelderMeadSimplex}} and {{MultiDirectionalSimplex}}.
>  * {{MultiDirectionalSimplex}} fails 2 of them.
>  * {{NelderMeadSimplex}} fails 10 of them.
> All failing tests are set to "@Ignore" for now. 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to