[
https://issues.apache.org/jira/browse/MATH-1108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13920658#comment-13920658
]
Thomas Neidhart commented on MATH-1108:
---------------------------------------
The problem with your original problem definition is that all the right hand
sides for the non-bounds constraints are 0.
This means that the min ratio test as implemented in our solver (and which is
the standard one) will have troubles finding the best way to improve the
solution as all are equally likely. The result is a very slow convergence.
Other solvers might handle this special case better, and we should also aim to
improve here. Could you tell me with which solvers you tried your problem and
got better results?
> SimplexSolver returns TooManyIterationsException
> ------------------------------------------------
>
> Key: MATH-1108
> URL: https://issues.apache.org/jira/browse/MATH-1108
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.2, 3.1.1
> Reporter: Buharov Alexander
> Attachments: Problem.zip
>
>
> We have vector v and matrix s (from file S.txt)
> v - vector ( size: 1*649 )
> s - matrix ( size: 554*649 )
> We want solve linear problem:
> * max (v[31])
> * Bounds:
> * s*v = 0
> * -1 <= v[i] <= 1 for i=0,..., 648
> We changed MaxIter parameter (from 100 to 1e+6). And we changed epsilon (
> set default value and 1e-3). But solver always throws
> TooManyIterationsException (and work very slow).
--
This message was sent by Atlassian JIRA
(v6.2#6252)
