[Couenne] 回复: Re: Re: problems in using couenne
Pietro Belotti
pbelott at clemson.edu
Sat Oct 9 11:16:28 EDT 2010
Dear Xu,
> at the position "<---", it said that "1e+050 best solution, best
> possible -17.1429 ". Does it mean Couenne has found out a solution with
> objective value -17.1429, while the time expended is 9966.95 seconds?
No, the -17.1429 is a lower bound, i.e., an estimate from below on the
value of the optimal solution, but it doesn't necessarily correspond to
any feasible solution. From your output it appears that no feasible
solution was found: you can tell this from the upper bound value of 1e+50.
> I hope Couenne to stop here and write the solution. I have tried
> couenne.opt file with "time_limit 12000" or "art_cutoff -17", and indeed
> Couenne would stop and write a sol file, but sol file did not have the
> value of objective and variables. How will I do, if I hope Couenne to
> stop at that time and write the solution?
The option "art_cutoff -17" is correct, however, if you don't know of any
feasible solution with value -17 or lower, you may end up eliminating all
feasible solutions. For a start, I would suggest using a larger value.
Again, you have to wait for Couenne to output "Integer solution found..."
before you can claim to have a feasible solution and to have it memorized
in a .sol file.
Also, as I said in my previous message, I believe the best way to solve
that problem is through a mixed integer *linear* model, rather than a
nonlinear one.
Hope this helps.
Pietro
--
Pietro Belotti
Dept. of Mathematical Sciences
Clemson University
email: pbelott at clemson.edu
phone: 864-656-6765
www: myweb.clemson.edu/~pbelott
On Sat, 9 Oct 2010, xzxnpu wrote:
> Dear Dr. Pietro
> Thanks for your reply and suggestion.
> From the output of Couenne, as follows:
>
>
> couenne:
> ANALYSIS TEST: Problem size before reformulation: 24648 variables (3309 integer)
> , 9486 constraints.
> Couenne: reformulation time 37.422s
> Problem size after reformulation: 25908 variables (3146 integer), 2308 constrai
> nts.
> ******************************************************************************
> This program contains Ipopt, a library for large-scale nonlinear optimization.
> Ipopt is released as open source code under the Common Public License (CPL).
> For more information visit http://projects.coin-or.org/Ipopt
> ******************************************************************************
> NOTE: You are using Ipopt by default with the MUMPS linear solver.
> Other linear solvers might be more efficient (see Ipopt documentation).
>
> 0 :Error allocating IS:IERR= 41
> MUMPS returned INFO(1) =-13 - out or memory.
> In some cases it helps to decrease the value of the option "mumps_mem_percent".
> 0 :Error allocating IS:IERR= 41
> MUMPS returned INFO(1) =-13 - out or memory.
> In some cases it helps to decrease the value of the option "mumps_mem_percent".
> Restoration phase is called at point that is almost feasible,
> with constraint violation 4.783227e-016. Abort.
> 0 :Error allocating IS:IERR= 41
> MUMPS returned INFO(1) =-13 - out or memory.
> In some cases it helps to decrease the value of the option "mumps_mem_percent".
> 0 :Error allocating IS:IERR= 41
> MUMPS returned INFO(1) =-13 - out or memory.
> In some cases it helps to decrease the value of the option "mumps_mem_percent".
> Restoration phase is called at point that is almost feasible,
> with constraint violation 4.783227e-016. Abort.
> Restoration phase is called at point that is almost feasible,
> with constraint violation 4.218932e-015. Abort.
> Cbc0031I 3 added rows had average density of 2
> Cbc0013I At root node, 3 cuts changed objective from -17.1429 to -17.1429 in 4 p
> asses
> Cbc0014I Cut generator 0 (Couenne convexifier cuts) - 3 row cuts average 2.0 ele
> ments, 759 column cuts (762 active)
> Cbc0010I After 0 nodes, 0 on tree, 1e+050 best solution, best possible -17.1429
> (9966.95 seconds) (<--- I hope to stop here and write solution file)
> Restoration phase is called at point that is almost feasible,
> with constraint violation 3.289202e-019. Abort.
> Restoration phase is called at point that is almost feasible,
> with constraint violation 3.027590e-015. Abort.
> Cbc0010I After 100 nodes, 25 on tree, 1e+050 best solution, best possible -17.14
> 29 (50115.01 seconds)
>
>
> at the position "<---", it said that "1e+050 best solution, best possible -17.1429 ".
> Does it mean Couenne has found out a solution with objective value -17.1429, while
> the time expended is 9966.95 seconds? I hope Couenne to stop here and write the solution.
> I have tried couenne.opt file with "time_limit 12000" or "art_cutoff -17", and indeed Couenne
> would stop and write a sol file, but sol file did not have the value of objective and variables.
> How will I do, if I hope Couenne to stop at that time and write the solution?
>
> 2010-10-09
>
> _________________________________________________________________________________________________________________________________________________________________________________
> xu
>
>
> _________________________________________________________________________________________________________________________________________________________________________________
> 发件人: Pietro Belotti <pbelott at clemson.edu>
> 发送时间: 2010-09-29 11:58
> 主 题: Re: Re: [Couenne] problems in using couenne
> 收件人: xzxnpu <xzxnpu at sina.com>
> 抄 送: couenne at list.coin-or.org
>
>
>
>
> Dear Xu,
>
> first, I would appreciate if you send your comments or questions to the
> Couenne mailing list, couenne at list.coin-or.org, so that the responses can
> be useful to other users of Couenne.
>
> Second, after a quick look at the model you have sent, it seems that your
> model can be easily converted into a mixed integer LINEAR programming
> (MILP) model, rather than a mixed integer nonlinear programming (MINLP)
> one. One advantage of this is that MILP problems can be handled by faster
> solvers, especially in a model like yours that admits tens of thousands of
> variables/constraints.
>
> In your model, the only nonlinear part is in the last constraint: it is a
> sum of terms x[j]*y[i], where x[i] is a binary variable and y[i] is
> between 0 and M, where M=10^10. You can obtain a similar result by writing
> the same sum with only the y[i] terms, and then adding the constraint
>
> y[i] <= M * x[j]
>
> for all pairs (i,j) that defined terms x[j]*y[i] in the sum at the last
> constraint. You get a MILP model that you can solve with Cbc, which
> is another Coin-OR package.
>
> > For my first question: I followed your suggestion to create a file
> > couenne.opt with “time_limit 12000”. Indeed Couenne would stop at the
> > time after running 12000 secconds, but it did not write a solution file
> > with the values of objective and variables (I have attached the sol file
> > that it generated, with name “1.sol”).
>
> Based on the output file you sent, after that time limit Couenne did not
> find any feasible solution: the upper bound is 10^50. The -17 is a lower
> bound and does not correspond to any feasible solution, which is therefore
> not saved.
>
> > I have tried another parameter “art_lower”, for I know the objective of
> > my MINLP model is about -17.1429, so I set “art_lower -17.1” and hope
> > Couenne to stop and write the solution, but it did not stop. I have
> > attached the output of Couenne in the two computation instances (no any
> > option parameter set / time_limit = 12000, with name “2.txt”) and give
> > the position where I hope Couenne to stop and write the solution at that
> > time. How to do? How about to use “couenne_check”?
>
> The parameter art_lower is used to set an artificial lower bound, and is
> in general not useful outside debugging purposes. If you know of a
> FEASIBLE solution whose value is, for example, 12345.0, I would suggest
> specifying an option
>
> art_cutoff 12345.0
>
> which instructs Couenne to ignore all solutions with value larger than
> that and may help finding an optimal solution. Notice that for this to
> help there must be a solution with that value. If you know the value of
> the x and y variables corresponding to that feasible solution, consider
> using the "default" keyword in AMPL, which would store that solution and
> allow Couenne to use it as an upper bound.
>
> > For my second question: Indeed, Couenne will write a sol file as you
> > said. But why do Ipopt, a solver for NLP and you know surely, not write
> > a sol file after it complete the computation for a model? How to make it
> > write while not using AMPL or other interfaces? Of course, Ipopt is not
> > Couenne. If you know the answer, please tell me.
>
> I believe there should be an option to instruct Ipopt to write a solution,
> if any exists. I do not know Ipopt's options, so I'd refer you to Ipopt's
> lists and web pages.
>
> > For my third question: I have attached the instance that generated this
> > problem with the names “3.nl” and “3.mod” (for AMPL). The version of
> > Couenne is binary for windows (“Couenne-0.3.2-win32-msvc9.zip”,
> > downloaded from http://www.coin-or.org/download/binary/Couenne/); the
> > type of machine is “Intel(R) Core(TM)2 Duo CPU” with Microsoft Windows
> > XP Professional (Service Pack 3).
>
> The problem seems to be related to a reformulation bug, which is already
> in the list of bugs I'm after.
>
> > For my forth question: The version of Couenne is binary for Linux
> > (Couenne-0.1.0-linux-x86_64-gcc4.3.2.tgz, also downloaded from
> > http://www.coin-or.org/download/binary/Couenne/ ). the type of machine
> > “Intel(R) Xeon(R) CPU” is with CentOS 5.5 and gcc version is less than
> > 4.3.0
>
> I would suggest downloading and compiling a more recent version of Couenne
> (see https://projects.coin-or.org/Couenne for instructions). Apart from
> giving you a more efficient solver, it may help configure and build an
> executable that knows where the Fortran libraries are.
>
> Hope this helps.
>
> Best,
> Pietro
>
> --
> Pietro Belotti
> Dept. of Mathematical Sciences
> Clemson University
> email: pbelott at clemson.edu
> phone: 864-656-6765
> www: myweb.clemson.edu/~pbelott
>
> On Wed, 29 Sep 2010, xzxnpu wrote:
>
> >
> > Dear Dr. Pietro
> >
> > Thanks for your reply.
> >
> > For my first question: I followed your suggestion to create a file couenne.opt with “time_limit 12000”. Indeed Couenne would stop at the time after running 12000 secconds, b
> ut
> > it did not write a solution file with the values of objective and variables (I have attached the sol file that it generated, with name “1.sol”). I have tried another paramete
> r
> > “art_lower”, for I know the objective of my MINLP model is about -17.1429, so I set “art_lower -17.1” and hope Couenne to stop and write the solution, but it did not stop. I
?? ?have
> > attached the output of Couenne in the two computation instances (no any option parameter set / time_limit = 12000, with name “2.txt”) and give the position where I hope Couen
> ne
> > to stop and write the solution at that time. How to do? How about to use “couenne_check”?
> >
> >
> >
> > For my second question: Indeed, Couenne will write a sol file as you said. But why do Ipopt, a solver for NLP and you know surely, not write a sol file after it complete the
>
> > computation for a model? How to make it write while not using AMPL or other interfaces? Of course, Ipopt is not Couenne. If you know the answer, please tell me.
> >
> >
> >
> > For my third question: I have attached the instance that generated this problem with the names “3.nl” and “3.mod” (for AMPL). The version of Couenne is binary for windows
> > (“Couenne-0.3.2-win32-msvc9.zip”, downloaded from http://www.coin-or.org/download/binary/Couenne/); the type of machine is “Intel(R) Core(TM)2 Duo CPU” with Microsoft Windows
> XP
> > Professional (Service Pack 3).
> >
> > For my forth question: The version of Couenne is binary for Linux (Couenne-0.1.0-linux-x86_64-gcc4.3.2.tgz, also downloaded from http://www.coin-or.org/download/binary/Couen
> ne/
> > ). the type of machine “Intel(R) Xeon(R) CPU” is with CentOS 5.5 and gcc version is less than 4.3.0
> >
> >
> >
> > Xu
> >
> > from CAS in china
> >
> >
> > 2010-09-29
> >
> > ______________________________________________________________________________________________________________________________________________________________________________
> ____
> > xzxnpu
> > ---------------------------------------------------------
> > 一键发送本地文件，请用网易闪电邮（fm.163.com）！
> >
> > ______________________________________________________________________________________________________________________________________________________________________________
> ____
> > 发件人: Pietro Belotti <pbelott at clemson.edu>
> > 发送时间: 2010-09-28 05:01
> > 主 题: Re: [Couenne] problems in using couenne
> > 收件人: xzx21c at 163.com
> > 抄 送: couenne at list.coin-or.org
> >
> >
> >
> >
> > Hello,
> >
> > > 1) I have used couenne, on AMPL, to solve one of my models which I sent
> > > to you previously. But the computation time is very long, more than a week,
> > > and now do not end. I hope to stop it while I hope to keep the computational
> > > solution for maybe couenne has found the optimal solution. How will I do? In
> > > the later, if I will make couenne to compute in a not
> > > long time, which option parameters should I set for couenne?
> >
> > create a file couenne.opt, which has to be in the same directory where you
> > run AMPL. The file should contain a line like the following
> >
> > time_limit 7200
> >
> > where the value of the parameter expresses the number of seconds after
> > which Couenne should stop if it didn't find an optimal solution yet.
> >
> > > 2) Now I run couenne on AMPL. I hope to know: if I just use couenne to
> > > solve a nl file (using “couenne *.nl” command, not on the platforms such as
> > > AMPL or GAMS), how to know and display the solution results?
> >
> > Please check
> >
> > http://list.coin-or.org/pipermail/couenne/2009-July/000041.html
> >
> > If the file is MyFolder/MyInstance.nl, running the command "couenne
> > MyFolder/MyInstance.nl" will generate a file MyFolder/MyInstance.sol. The
> > last n lines of the .sol file contain the value of the variables in the
> > best solution found by Couenne.
> >
> > > 3) I have used couenne of windows version, but it can run just on one
> > > computer and can not run on other windows computers. If the computer have
> > > installed visual studio, the error messages will be as
> > >
> > > An unhandled win32 exception occurred in couenne.exe [2660].
> > >
> > > Possible debuggers: New instance of Visual Studio 2008.
> > >
> > > Do you want to debug using the selected debugger?
> > >
> > > If the computer have not installed visual studio, couenne will stop and
> > > the error messages will be as
> > >
> > > An unhandled win32 exception occurred in couenne.exe
> > >
> > > I don’t know what is the problem?
> >
> > I did not encounter such problem before. It might be useful if you send
> > the instance that generated this problem and provide more information:
> > the version of Couenne you have been using, the type of machine, the
> > compiler version, etc.
> >
> > > 4) Couenne can not be run on my linux computer. When I run it, messages
> > > as 'error while loading shared libraries: libgfortran.so.3: cannot open
> > > shared object file: No such file or directory' will appear. I have doubted my
> > > gcc version, so I tried several couenne version, similar message will appear
> > > but the shared libraries libgfortran.so.* are different.
> > > I do not know what is the problem as well?
> >
> > That might depend strongly on the configuration. According to a recent
> > post,
> >
> > http://list.coin-or.org/pipermail/couenne/2010-September/000123.html
> >
> > the following configuration option might help:
> >
> > $ ../configure FFLAGS="-m64 -O3 -fomit-frame-pointer -pipe"
> >
> > Let me know if this helps.
> >
> > Regards,
> > Pietro
> >
> >
> > --
> > Pietro Belotti
> > Dept. of Mathematical Sciences
> > Clemson University
> > email: pbelott at clemson.edu
> > phone: 864-656-6765
> > www: myweb.clemson.edu/~pbelott
> >
> > On Mon, 27 Sep 2010, Pietro Belotti wrote:
> >
> > >
> > > hello!
> > >
> > > I have encountered some problems in using couenne.
> > >
> > > 1) I have used couenne, on AMPL, to solve one of my models which I sent
> > > to you previously. But the computation time is very long, more than a week,
> > > and now do not end. I hope to stop it while I hope to keep the computational
> > > solution for maybe couenne has found the optimal solution. How will I do? In
> > > the later, if I will make couenne to compute in a not
> > > long time, which option parameters should I set for couenne?
> > >
> > > 2) Now I run couenne on AMPL. I hope to know: if I just use couenne to
> > > solve a nl file (using “couenne *.nl” command, not on the platforms such as
> > > AMPL or GAMS), how to know and display the solution results?
> > >
> > > 3) I have used couenne of windows version, but it can run just on one
> > > computer and can not run on other windows computers. If the computer have
> > > installed visual studio, the error messages will be as
> > >
> > > An unhandled win32 exception occurred in couenne.exe [2660].
> > >
> > > Possible debuggers: New instance of Visual Studio 2008.
> > >
> > > Do you want to debug using the selected debugger?
> > >
> > > If the computer have not installed visual studio, couenne will stop and
> > > the error messages will be as
> > >
> > > An unhandled win32 exception occurred in couenne.exe
> > >
> > > I don’t know what is the problem?
> > >
> > > 4) Couenne can not be run on my linux computer. When I run it, messages
> > > as 'error while loading shared libraries: libgfortran.so.3: cannot open
> > > shared object file: No such file or directory' will appear. I have doubted my
> > > gcc version, so I tried several couenne version, similar message will appear
> > > but the shared libraries libgfortran.so.* are different.
> > > I do not know what is the problem as well?
> >
> > _______________________________________________
> > Couenne mailing list
> > Couenne at list.coin-or.org
> > http://list.coin-or.org/mailman/listinfo/couenne
> >
> >
>
>
More information about the Couenne
mailing list