[Cbc] [Clp] Clp segmentation fault

[Matthew Gidden]

Hi Matthew,


On Tue, Jul 7, 2015 at 10:09 PM, Matthew Bromberg <mcbromberg at gmail.com>
wrote:

> I am interfacing to Clp through the python package cylp. Unfortunately it
> seems to only support cbc-2.8.5.


I was just able to compile cylp agaisnt the current trunk. Have you given
that (or similar) a try and failed? According to the install file, it looks
for coin libraries in either your COIN_INSTALL_PATH (if set) or the result
of `which clp` (for me this is my custom install location.

$ which clp
> /home/gidden/.local/bin/clp





> It is a large scale linear programming problem whose constraint matrix is
> of dimension 40K x 10K.
>
> Here is what happens when I attempt to optimize this model:
>
>  $ python mycode.py
> Clp0006I 0  Obj 0 Dual inf 13.083708 (9998) w.o. free dual inf (0)
> Clp0006I 12  Obj -0.59273256 Dual inf 494.36229 (9979) w.o. free dual inf
> (5)
> Clp0006I 91  Obj -1.2553307 Dual inf 60372.348 (9862) w.o. free dual inf
> (44)
> Clp0006I 291  Obj -2.369027 Dual inf 4562.9727 (9559) w.o. free dual inf
> (141)
> Clp0006I 491  Obj -3.492329 Dual inf 4602.3626 (9264) w.o. free dual inf
> (246)
> Clp0006I 691  Obj -4.4089791 Dual inf 2491.8024 (8955) w.o. free dual inf
> (337)
> Segmentation fault
>
> I'm expecting it to find a solution around -100 or so,  so it segfaulted
> rather early.  I'd like to try it against cbc-2.9.5, but I have no idea how
> to do that.
> Any MPS file created for this would be over 100 GBytes in size, so that's
> ruled out as an input format.  My data is stored in Matlabs hdf5 format
> (matlab 7.3).
>
> How does Clp handle a subset of constraints that is nearly singular?
>
> By the way I've been pleasantly surprised at how fast this code runs on
> smaller problems being more memory efficient apparently than some
> commercial products.  It's also surprising considering that it is not
> really effectively parallelized.  Probably no more than the underlying
> multithreaded Blas libraries.
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc
>



-- 
Matthew Gidden, Ph.D.
Postdoctoral Associate, Nuclear Engineering
The University of Wisconsin -- Madison
Ph. 225.892.3192
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20150708/d9f6fa6c/attachment.html>