[SCIP] SCIP and Ipopt
Ambros Gleixner
gleixner at zib.de
Wed Feb 21 20:25:46 CET 2018
>> Also, I would like to know if it is possible to pause SCIP and
>> retrieve its
>> state and, then, continue from the saved state from the
>> interactive/batch(script) mode. If so, how?
>
> No.
Well, I would say you can do that to some extent by setting any
parameter that limits the effort spent for solving, e.g., limits/nodes,
limits/solutions, limits/time, ...
For continuing after SCIP stopped, simply increase the limit and type
optimize again.
Best,
Ambros
Am 21.02.2018 um 11:41 schrieb Stefan Vigerske:
> Hi,
>
> On 02/21/2018 12:16 AM, Vincent Mirian wrote:
>> Hi Ambros.
>>
>> Thank you for your response. Your explanation was aligned with my
>> assumption... Thank you for clarifying.
>>
>> SCIP searches for an optimal solution. For some MINLP problem, I would
>> like
>> SCIP to return a solution (the first solution it may find is
>> sufficient). I
>> was reading that IPOPT is integrated within SCIP, however I was unable to
>> trigger ipopt for these problem. Any help would be appreciated.
>
> There are a few primal heuristics in SCIP that may call out to Ipopt.
> One that runs rather frequently is the "subNLP" heuristic, but it will
> not run when there are no continuous variables in the nonlinear part of
> your problem. Otherwise, you may change its parameters to make it run
> more often.
> However, with the information given, it's hard to say why SCIP didn't
> call Ipopt.
>
>> For reference, I am using SCIP from interactive/batch(script) mode. I
>> installed SCIP using the binaries posted on the website.
>>
>> Also, I would like to know if it is possible to pause SCIP and
>> retrieve its
>> state and, then, continue from the saved state from the
>> interactive/batch(script) mode. If so, how?
>
> No.
>
> Stefan
>
>>
>> Thank you in advance.
>>
>> On Tue, Feb 20, 2018 at 1:49 AM, Ambros Gleixner <gleixner at zib.de> wrote:
>>
>>> Hi Vincent,
>>>
>>> SCIP (like virtually all solvers based on floating-point arithmetics)
>>> uses
>>> tolerances to check feasibility and integrality. Anything with a
>>> fractionality below 1e-9 (parameter numerics/epsilon) is consider
>>> integral,
>>> so also the value 0.999999999999998 that you saw in your solution.
>>> You can
>>> use SCIPisIntegral() to check or SCIPround() to round solution values in
>>> the C API.
>>>
>>> Best,
>>> Ambros
>>>
>>>
>>> Am 20.02.2018 um 03:40 schrieb Vincent Mirian:
>>>
>>>> Hi Benny.
>>>>
>>>> Thank you for your response. Currently, I am using the LP format with
>>>> SCIP.
>>>>
>>>> I create a problem (attached). In the problem I defined a
>>>> variable AssignC4BLK0 and AssignC10BLK1 as binaries. However, the
>>>> output
>>>> assigns a non-binary to these variables. Is there something that I
>>>> am doing
>>>> incorrectly? Or is this natural behavior from SCIP?
>>>>
>>>> Regards.
>>>>
>>>> On Wed, Feb 14, 2018 at 1:55 AM, Benjamin Müller
>>>> <benjamin.mueller at zib.de
>>>> <mailto:benjamin.mueller at zib.de>> wrote:
>>>>
>>>> Hi Vincent,
>>>>
>>>> you did two mistakes:
>>>>
>>>> 1. <y#0> should have a coefficient of 1 in the objective.
>>>>
>>>> 2. You declared your quadratic constraint to be linear and there
>>>> were also some semantic problems.
>>>>
>>>> Your model should actually look like:
>>>>
>>>> STATISTICS
>>>> Problem name : test.cip
>>>> Variables : 5 (0 binary, 5 integer, 0 implicit
>>>> integer, 0
>>>> continuous)
>>>> Constraints : 0 initial, 5 maximal
>>>> OBJECTIVE
>>>> Sense : minimize
>>>> VARIABLES
>>>> [integer] <c#0b#0>: obj=0, original bounds=[0,1]
>>>> [integer] <c#1b#0>: obj=0, original bounds=[0,1]
>>>> [integer] <c#0b#1>: obj=0, original bounds=[0,1]
>>>> [integer] <c#1b#1>: obj=0, original bounds=[0,1]
>>>> [integer] <y#0>: obj=1, global bounds=[0,4], local bounds=[0,4]
>>>> CONSTRAINTS
>>>> [linear] <pack_1>: <c#0b#0>[I] +<c#0b#1>[I] == 1;
>>>> [linear] <pack_2>: <c#1b#0>[I] +<c#1b#1>[I] == 1;
>>>> [linear] <cap_0>: <c#0b#0>[I] +<c#1b#0>[I] <= 1;
>>>> [linear] <cap_1>: <c#0b#1>[I] +<c#1b#1>[I] <= 1;
>>>> [quadratic] <obj_0>: -<c#0b#1>[C] <c#1b#1>[C] -<c#0b#0>[I]
>>>> <c#1b#0>[I] -<y#0>[C] == -4;
>>>> END
>>>>
>>>> Please read your instance with SCIP display the problem in
>>>> order to
>>>> see how SCIP parsed it.
>>>>
>>>> ./bin/scip -c "read test.cip display problem"
>>>>
>>>> Anyway, I think that you might want to use a modelling language
>>>> instead of dealing with our .cip format. You could, e.g., use
>>>> ZIMPL,
>>>> AMPL, GAMS, ...
>>>>
>>>> Best,
>>>> Benny
>>>>
>>>> PS: Please don't include the Ipopt mailing any more. Your problems
>>>> are not related to Ipopt.
>>>>
>>>>
>>>>
>>>> On 02/14/2018 07:01 AM, Vincent Mirian wrote:
>>>>
>>>> Hi all.
>>>>
>>>> Thanks for the reply Benny.
>>>>
>>>> I created the following input file to the framework.
>>>> STATISTICS
>>>> Problem name : test.cip
>>>> Variables : 5 (0 binary, 5 integer, 0 implicit
>>>> integer, 0 continuous)
>>>> Constraints : 0 initial, 5 maximal
>>>> OBJECTIVE
>>>> Sense : minimize
>>>> VARIABLES
>>>> [integer] <c#0b#0>: obj=0, original bounds=[0,1]
>>>> [integer] <c#1b#0>: obj=0, original bounds=[0,1]
>>>> [integer] <c#0b#1>: obj=0, original bounds=[0,1]
>>>> [integer] <c#1b#1>: obj=0, original bounds=[0,1]
>>>> [integer] <y#0>: obj=0, global bounds=[0,4], local
>>>> bounds=[0,4]
>>>> CONSTRAINTS
>>>> [linear] <pack_1>: <c#0b#0>[I] +<c#0b#1>[I] == 1;
>>>> [linear] <pack_2>: <c#1b#0>[I] +<c#1b#1>[I] == 1;
>>>> [linear] <cap_0>: <c#0b#0>[I] +<c#1b#0>[I] <= 1;
>>>> [linear] <cap_1>: <c#0b#1>[I] +<c#1b#1>[I] <= 1;
>>>> [linear] <obj_0>: -<c#0b#1>[I]*<c#1b#1>[I]*1
>>>> -<c#0b#0>[I]*<c#1b#0>[I]*1 +4 -<y#0> == 0;
>>>> END
>>>>
>>>> The results should assign y#0 to 4. However, I get the
>>>> following
>>>> output (snippet of output to reduce length):
>>>>
>>>> presolving (1 rounds: 1 fast, 1 medium, 1 exhaustive):
>>>> 5 deleted vars, 5 deleted constraints, 0 added
>>>> constraints, 1
>>>> tightened bounds, 0 added holes, 0 changed sides, 0 changed
>>>> coefficients
>>>> 0 implications, 0 cliques
>>>> presolved problem has 0 variables (0 bin, 0 int, 0 impl, 0
>>>> cont)
>>>> and 0 constraints
>>>> transformed objective value is always integral (scale: 1)
>>>> Presolving Time: 0.00
>>>>
>>>> objective value: 0
>>>> c#0b#0 1 (obj:0)
>>>> c#1b#1 1 (obj:0)
>>>>
>>>> The assignment of c#0b#0 and c#1b#1 are valid solutions. I am
>>>> confused on the reason that the objective value is 0. I am
>>>> assume that the objective value is y#0, since it is the only
>>>> variable not stated. The value of y#0 should be 4.
>>>>
>>>> Some guidance would be appreciated. Note that I searched the
>>>> SCIP mailing list archive but I did not find any references
>>>> for
>>>> the input format and description of the output.
>>>>
>>>> Any help would be appreciated. Thank you.
>>>>
>>>>
>>>>
>>>> On Tue, Feb 13, 2018 at 11:38 PM, Benjamin Müller
>>>> <benjamin.mueller at zib.de <mailto:benjamin.mueller at zib.de>
>>>> <mailto:benjamin.mueller at zib.de
>>>>
>>>> <mailto:benjamin.mueller at zib.de>>> wrote:
>>>>
>>>> Hi Vincent,
>>>>
>>>> I would recommend using JSCIPOpt, which is our SCIP-Java
>>>> interface
>>>> that can be found here:
>>>>
>>>> https://github.com/SCIP-Interfaces/JSCIPOpt
>>>> <https://github.com/SCIP-Interfaces/JSCIPOpt>
>>>> <https://github.com/SCIP-Interfaces/JSCIPOpt
>>>> <https://github.com/SCIP-Interfaces/JSCIPOpt>>
>>>>
>>>> All relevant steps for building the interface are
>>>> listed in
>>>> the
>>>> INSTALL.md, but basically, you will need to download the
>>>> latest SCIP
>>>> Optimization Suite and compile it with Ipopt. You might
>>>> want to have
>>>> a look at the JSCIPOpt/examples/Quadratic, which shows
>>>> how
>>>> to model
>>>> a problem with simple quadratic constraints.
>>>>
>>>> Please note that SCIP can't handle a nonlinear objective
>>>> function.
>>>> For this reason, you need to add an auxiliary variable z
>>>> and then
>>>> model your problem as
>>>>
>>>> min z
>>>> z >= sum_{ij} X_{ij} Y_{ij}
>>>> ...
>>>>
>>>> Best,
>>>> Benny
>>>>
>>>> On 02/13/2018 07:05 PM, Vincent Mirian wrote:
>>>>
>>>> HI all.
>>>>
>>>> I am new to the SCIP community.
>>>>
>>>> I am looking for a non-linear solver to solve:
>>>> - an objective function similar to 0 <= sum over i
>>>> and
>>>> j of X_ij
>>>> * Y_ij * Constant < max
>>>> - with constraints similar tosum(X_ij) = 1 and
>>>> sum(Y_ij) = 1
>>>> (forcing the values of X_ij and Y_ij to zero or
>>>> one). A
>>>> solution
>>>> would be sufficient (it can be local or global
>>>> optimal).
>>>>
>>>> I would need to interface the solver with Java. I
>>>> found
>>>> Google
>>>> Optimization Tools (OR-Tools) to interface SCIP with
>>>> java. Note
>>>> that this work is for academic research.
>>>>
>>>> - Would SCIP or Ipopt be suitable for my task? Would
>>>> anyone have
>>>> experience integrating SCIP or Ipopt into Java?
>>>> - I'm using a machine with ubuntu 64-bit. Are
>>>> there any
>>>> issues
>>>> with installing these tools in this machine
>>>> environment?
>>>> - What is the manner to describe the constraints and
>>>> objective
>>>> function using these tools?
>>>>
>>>> Thank you.
>>>> -- Vincent Mirian
>>>>
>>>>
>>>> _______________________________________________
>>>> Scip mailing list
>>>> Scip at zib.de <mailto:Scip at zib.de> <mailto:Scip at zib.de
>>>> <mailto:Scip at zib.de>>
>>>> https://listserv.zib.de/mailman/listinfo/scip
>>>> <https://listserv.zib.de/mailman/listinfo/scip>
>>>> <https://listserv.zib.de/mailman/listinfo/scip
>>>> <https://listserv.zib.de/mailman/listinfo/scip>>
>>>>
>>>>
>>>> -- ______________________________
>>>> Benjamin Müller
>>>> Zuse Institute Berlin
>>>> Takustr. 7, 14195 Berlin
>>>> benjamin.mueller at zib.de <mailto:benjamin.mueller at zib.de>
>>>> <mailto:benjamin.mueller at zib.de
>>>> <mailto:benjamin.mueller at zib.de>>
>>>> +49 30 841 85-195 <tel:%2B49%2030%20841%2085-195>
>>>> <tel:%2B49%2030%20841%2085-195>
>>>>
>>>>
>>>>
>>>>
>>>> -- Vincent Mirian
>>>>
>>>>
>>>> -- ______________________________
>>>> Benjamin Müller
>>>> Zuse Institute Berlin
>>>> Takustr. 7, 14195 Berlin
>>>> benjamin.mueller at zib.de <mailto:benjamin.mueller at zib.de>
>>>> +49 30 841 85-195 <tel:%2B49%2030%20841%2085-195>
>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Vincent Mirian
>>>>
>>>>
>>>> _______________________________________________
>>>> Scip mailing list
>>>> Scip at zib.de
>>>> https://listserv.zib.de/mailman/listinfo/scip
>>>>
>>>>
>>> --
>>> Ambros Gleixner, Research Group Mathematical Optimization Methods at
>>> Zuse
>>> Institute Berlin, http://www.zib.de/gleixner
>>> _______________________________________________
>>> Scip mailing list
>>> Scip at zib.de
>>> https://listserv.zib.de/mailman/listinfo/scip
>>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> https://listserv.zib.de/mailman/listinfo/scip
>>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
--
Ambros Gleixner, Research Group Mathematical Optimization Methods at
Zuse Institute Berlin, http://www.zib.de/gleixner
More information about the Scip
mailing list