optional SCIP integer constraint solver (Ticket #10879)

[S]

Hi,

I noticed that Martin started work on a SCIP wrapper for SAGE.
http://trac.sagemath.org/sage_trac/ticket/10879

I wanted to add two other ideas for interfacing with SCIP that are
written in python:

Numberjack
http://4c110.ucc.ie/numberjack/

python-zibopt
http://code.google.com/p/python-zibopt/

I'd love to use SCIP through SAGE, but am a newbie with SAGE so I
don't know how to help out.

Best, Shawn

[Martin Albrecht]

On Thursday 21 April 2011, S wrote:
> Hi,

Hi,

> I noticed that Martin started work on a SCIP wrapper for SAGE.
> http://trac.sagemath.org/sage_trac/ticket/10879

> I wanted to add two other ideas for interfacing with SCIP that are
> written in python:
>
> Numberjack
> http://4c110.ucc.ie/numberjack/
>
> python-zibopt
> http://code.google.com/p/python-zibopt/

I knew python-zibopt but I didn't know Numberjack which looks interesting. The
reason I didn't go for python-zibopt is:

- we wanted a SCIP interface which matches our generic MIP interface such that
you can use SCIP wherever MIP is used in Sage (I could have written that on
top of python-zibopt I guess, but that would have meant more overhead)

- I wanted more constraints than python-zibopt offers. It only does MIP but I
want to access all the constrains SCIP has to offer (non-linear constraints
etc.)

- Finally, it was a good way for me to learn the API (and some internals) of
SCIP :)

> I'd love to use SCIP through SAGE, but am a newbie with SAGE so I
> don't know how to help out.

Sage is meant to be easy to get into in so let's test that :) What is your
interest? Do you want to add optional packages for Numberjack & python-zibopt
or are you happy with improving the current SCIP interface? If you want to
help out with the latter here are some ideas :)

- If you run Linux of some sort: just try it out, does it install okay for
you, does it work?
- If you run OSX and know your way around C you could help to fix the SIGSEGV
we see there
- You could read the documentation and improve it (i.e. provide documentation
patches for our SCIP interface)
- You could provide feedback what kind of interface would be nice to have,
i.e. how we can improve the interface

Cheers,
Martin

--
name: Martin Albrecht
_pgp: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8EF0DC99
_otr: 47F43D1A 5D68C36F 468BAEBA 640E8856 D7951CCF
_www: http://martinralbrecht.wordpress.com/
_jab: martinr...@jabber.ccc.de

[Harald Schilly]

On Thursday, April 21, 2011 7:03:55 AM UTC+2, S wrote:

python-zibopt
   http://code.google.com/p/python-zibopt/

It's only for scip 1.2, we target scip 2.x !

H

[S]

On Apr 21, 2:07 am, Martin Albrecht <martinralbre...@googlemail.com>
wrote:

> - I wanted more constraints than python-zibopt offers. It only does MIP but I
> want to access all the constrains SCIP has to offer (non-linear constraints
> etc.)

Yes, me too! I too want to take advantage of some of the constraint
programming aspects of SCIP, such as expressing ALLDIFFERENT
constraints, which is a real chore with pure MIP expressions.

> What is your interest?

Ultimately, I want to use SCIP and SAGE to answer Operations Research
questions at the school district that I work at.
I am happy to contribute to learn the API and to help SAGE become a
good math programming language. I have read up on PuLP and Coopr some
as well, but I really like how SAGE does so much more than numerical
optimization.

> Do you want to add optional packages for Numberjack & python-zibopt or are you happy with improving the current SCIP interface?

It would probably be easier for me to work on the interface.

> - If you run Linux of some sort: just try it out, does it install okay for you, does it work?

I will try this but have win32 OS, so I will have to do all this
through virtualbox -- is it possible to install SCIP this way?
(I can try and find out too.)

> - You could read the documentation and improve it (i.e. provide documentation patches for our SCIP interface)

This sounds good, probably at my level of current ability as a
newbie.
I could also convert some other problems from PuLP to SAGE syntax to
add other examples to one of the tutorial websites
http://www.sagemath.org/doc/reference/sage/numerical/mip.html
http://www.sagemath.org/doc/thematic_tutorials/linear_programming.html#

> - You could provide feedback what kind of interface would be nice to have, i.e. how we can improve the interface

Definitely, I'll let you know once I get started.

On a slightly different topic:
I would also be interested in adding MIP support for other solvers
such as Gurobi.
I have Gurobi, but not CPLEXl; therefore without SCIP, I'm out of luck
solving MIP problems where CBC is too slow or are quadratic in
nature. PuLP provides a nice interface directly to the Gurobi API
which we could learn from, or another route is that Gurobi also
accepts LP files which SAGE can already create. Here is a video from
Gurobi ( http://www.gurobi.com/flash/shell-2011-02.html ) and where
they show how to the same an LP file could be read into Gurobi.

Thanks,

Shawn

[S]

I think the developer of zipopt plans on targeting the new version,
but hasn't yet.

[Martin Albrecht]

> > > - If you run Linux of some sort: just try it out, does it install okay
> > for you, does it work?
>
> I will try this but have win32 OS, so I will have to do all this
> through virtualbox -- is it possible to install SCIP this way?
> (I can try and find out too.)

Yes, it should "just work" (TM) in your VirtualBox :)

> > - You could read the documentation and improve it (i.e. provide
> > documentation patches for our SCIP interface)
>
> This sounds good, probably at my level of current ability as a
> newbie.
> I could also convert some other problems from PuLP to SAGE syntax to
> add other examples to one of the tutorial websites
> http://www.sagemath.org/doc/reference/sage/numerical/mip.html
> http://www.sagemath.org/doc/thematic_tutorials/linear_programming.html#
>
> > - You could provide feedback what kind of interface would be nice to
> > have, i.e. how we can improve the interface
>
> Definitely, I'll let you know once I get started.
>
> On a slightly different topic:
> I would also be interested in adding MIP support for other solvers
> such as Gurobi.
> I have Gurobi, but not CPLEXl; therefore without SCIP, I'm out of luck
> solving MIP problems where CBC is too slow or are quadratic in
> nature. PuLP provides a nice interface directly to the Gurobi API
> which we could learn from, or another route is that Gurobi also
> accepts LP files which SAGE can already create. Here is a video from
> Gurobi ( http://www.gurobi.com/flash/shell-2011-02.html ) and where
> they show how to the same an LP file could be read into Gurobi.

I know Gurobi and IIRC there was some plan to add an interface to that too. I
think Harald Schilly did some work with the Gurobi interface but no-one worked
on a proper integration into Sage yet.

[Nathann Cohen]

I know Gurobi and IIRC there was some plan to add an interface to that too. I
think Harald Schilly did some work with the Gurobi interface but no-one worked
on a proper integration into Sage yet.

I tried that a dozen of times -- my main problem is that I have no way to obtain a license for Gurobi. One needs to have one's computer on a university network, and the research lab I work in is not recognised as such. It works by running a program on your own computer, this program connects to their servers and build a licence file using information on your local hardware. For this reason even using sage.math is useless...

After several emails with them, they still stuck to the same lines, and I gave it up....

Though if their authentication system were to change .... ^^;

Nathann

[S]

Nathann,

If your university does have access, but your building does not you
could connect to their network through something like Cisco VPN.

Shawn