[Cbc] [CBC] On closer inspection - solution discarded

[mg]

Sorry, I forgot to mention that I solve the 2 steps separately (I mean
creating and filling a new cbcmodel object at each step) and using callCbc()

2014-09-12 11:43 GMT+02:00 mg <giunto.cardanico at gmail.com>:

> Hi All,
> sorry if I'm bothering you lately but I hope this issues reporting will
> lead to a better and better version of CBC :)
>
> My problem is the following:
>
> I have a quite complex MIP model, that usually has some "acceptable"
> infeasibilities.
> So, I solve it in 2 steps :
>  -  first step : I add some acceptable slacks in some constraints and the
> objective is to minimize these slacks
>  -  second step : I modify the constraints right-hand-sides according to
> the slacks found in the previous step and run with the desired objective
> function.
>
> This usually works great , but sometimes it happens that the second step
> fails in founding a solution or results in an infeasible problem even if
> feasibility pump seemed to find a solution that is finally discarded with
> the error : "On closer inspection - solution discarded".
> Funny thing is, if I write the LP and re-run it in CBC, it works fine and
> the returned solution seems to be the one found and discarded in the
> previous run !
>
> I think this is due to some floating-point accuracy problems (got lost in
> the lp conversion), but I have no idea how can I solve it...
>
> Also, I would like to reproduce it and submit it to you, but it's quite
> complex, and, as I said, dumping the LP is not enough to reproduce the
> issue.
>
> At the moment I'm running on this code revisions :
> CBC : 2043
> CGL: 1028
> CLP : 2036
> CoinUtils : 1721
> (so it's almost the trunk version)
>
> As soon as I can I'm going to try to run it on the trunk version and
> report the result.
> In the meantime, do you have any suggestion ? For instance, some way to
> "increase" the tolerance in the second step ?
>
> Thanks in advance,
> Marco
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20140912/a1711129/attachment.html>