[SCIP] SCIP and Pyomo

[Vladimir V. Voloshinov]

Dear Salva,
let me a few comments...
1.

> Then I was wondering if moving to SCIP would entail any advantage. Based
> on your response, I would say there is no advantage.
>
There ARE advantage only if you want to find a global optimum (or, not
optimal, but feasible solution with "guaranteed gap" to unknown optimal
value), because SCIP implements Branch-and-bound algorithm for both
mixed-integer and "fully continuous" optimization problems.
2.

> ... on the advantages of generating NL-files myself
>
2.1. Generating of NL-file may be rather time consuming for problems of
rather high dimension. So, you can generate NL-file once and then try to
solve it with different solver options, or pass it to different solvers.
E.g. if nevertheless you want to find global optimum, after IPOPT, you can
pass NL-file to SCIPAMPL, or (after some conversion of file format) to pass
converted NL-file to multi-threaded FiberSCIP (for multi-core servers) or
parallel ParaSCIP (for MPI-clusters). More over, FiberSCIP or ParaSCIP
results may be converted to SOL-files to import solutions back to Pyomo.
2.2. In our researches we need in separation of NL-file generation and
solvers running on various computing resources of our distributed
environment https://optmod.distcomp.org/about. As to Pyomo you can see
https://github.com/distcomp/pyomo-everest and
https://optmod.distcomp.org/apps/vladimirv/solve-set-opt-probs   (please,
do not treat as an advertisement ;O)).
Best regards,
Vladimir.

On Mon, May 25, 2020 at 6:14 PM Salvador Pineda <salvapineda2 at gmail.com>
wrote:

> Dear Vladimir,
> Thanks a lot for your reply.
> Yes, I'm just interested in a local optimal solution to a continuous
> non-convex problem. Then I was wondering if moving to SCIP would entail any
> advantage. Based on your response, I would say there is no advantage.
> I am not sure either I completely understand the advantage of generating
> myself the NL-files. In fact, I use Pyomo because to avoid that. But
> probably there is something I am not taken into account. Would you
> elaborate on the advantages of generating NL-files myself?
> Best,
> Salva.
>
> On Mon, May 25, 2020 at 4:19 PM Vladimir V. Voloshinov <
> vladimir.voloshinov at gmail.com> wrote:
>
>>
>> Dear Salva,
>>
>> if there are no any discrete variables in your problem and you do not
>> want to treat it as a global optimization one, e.g. it is convex, or
>> unimodal (there is only one optimal solution), and local optimum will be
>> enough for you, then, you can use IPOPT "directly" from Pyomo. There is
>> special Pyomo-IPOPT interface...
>> But, frankly speaking, it is much more convenient to use IPOPT by the
>> same scheme: Pyomo->NL-file->IPOPT->SOL-file->Pyomo. E.g. you can generate
>> NL-files on your "small computer" and send them to a power server with
>> IPOPT (may be run in multi-threaded  mode on multi-core system) for solving
>> and return SOL-files  back for processing results via Python.
>>
>> Vladimir.
>>
>> On Mon, May 25, 2020 at 4:40 PM Salvador Pineda <spinedamorente at gmail.com>
>> wrote:
>>
>>> Thanks a lot Vladimir,
>>> But then, what is the advantage of using IPOPT under SCIP with PYOMO
>>> versus using directly IPOPT in PYOMO?
>>> Salva.
>>>
>>> On Mon, May 25, 2020 at 2:35 PM Vladimir V. Voloshinov <
>>> vladimir.voloshinov at gmail.com> wrote:
>>>
>>>> Dear Salvador,
>>>> answering your questions (basing on our experience of SCIP usage) :
>>>> 1) Yes, SCIP does use IPOPT for NLP, but, theoretically other
>>>> NLP-solvers may be used via NLP API
>>>> <https://scip.zib.de/doc-7.0.0/html/group__NLPIS.php>
>>>> 2) Our team successfully use SCIPAMPL (AMPL-compatible SCIP-version) by
>>>> Pyomo: Pyomo makes AMPL NL-file as SCIPAMPL-input; SCIPAMPL returns AMPL
>>>> SOL-file as output; Pyomo reads SOL-file.
>>>> It is important that above scheme may be used for ParaSCIP & FiberSCIP
>>>> with some additional trick...
>>>>
>>>> Yours sincerely,
>>>> Vladimir V. Voloshinov, web: GoogleScholar profile
>>>> <https://scholar.google.ru/citations?hl=en&user=-m4QhNEAAAAJ&view_op=list_works&sortby=pubdate>
>>>>
>>>> On Mon, May 25, 2020 at 2:06 PM Salvador Pineda <salvapineda2 at gmail.com>
>>>> wrote:
>>>>
>>>>> Hi,
>>>>> I have a couple of questions regarding SCIP solver.
>>>>> 1) Does SCIP rely on other solvers such as IPOPT to solve non-linear
>>>>> optimization problems?
>>>>> 2) Can SCIP be used under PYOMO?
>>>>> Thanks a lot!
>>>>> Salva.
>>>>> _______________________________________________
>>>>> Scip mailing list
>>>>> Scip at zib.de
>>>>> https://listserv.zib.de/mailman/listinfo/scip
>>>>>
>>>>
>>>>
>>>
>>> --------------------------------------------------------------------------
>>>
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20200526/9dedc295/attachment.html>