coin-or / dip

DIP is a decomposition-based solver framework for mixed integer linear programs.

DIP README

Welcome to the README for the Decomposition in Integer Programming (DIP) Framework. DIP is distributed under the Eclipse Public License and is freely redistributable. All source code and documentation is Copyright Matthew Galati, Ted Ralphs, Lehigh University, and others. This README may be redistributed freely.

Current Testing Status

Download and Install

Binaries for most platforms are available for download from Bintray. However, be aware that on Linux and OSX, you may have difficulties if you are using a distribution with different versions of various libraries than the build machine.

For DipPy, see Pypi.

For full installation instructions, see the INSTALL file.

Quickstart with DIP/DipPy

If you are on Linux or OS X, DIP can be installed from source using coinbrew. In Windows, there are pre-built binary wheels for DipPy and binaries on Bintray. for DIP. After build and install of DIP, if you are installing DipPy, make sure to set PKG_CONFIG_PATH to point to the directory where the .pc files are installed (by default, this is in the lib/pkgconfig directory in the installation directory). You may also need to set either LD_LIBRARY_PATH (Linux) or DYLD_LIBRARY_PATH (OS X) to point to the directory where the libraries are installed. Then simply do

pip install coinor.dippy

After installation, you can try running the examples to see if trhings are working. These should all work out of the box. For a listing of examples and some brief instructions, do

python -m coinor.dippy.examples

To get help for an example, do, e.g.,

python -m coinor.dippy.example.cflp --help

or just run

python -m coinor.dippy.examples.cflp

Cite

DOCUMENTATION

For a quick start guide, please see the INSTALL file in this distribution. Automatically generated documentation of the source code can be found here:

http://www.coin-or.org/Doxygen/Dip/

Further information can be found here:

http://github.com/coin-or/Dip

SUPPORT

List Serve

DIP users should either use the Dip mailing list or open an issue to ask questions to subscribe to the mailing list, go to http://list.coin-or.org/mailman/listinfo/dip

Bug Reports

Bug reports should be reported on github at

https://github.com/Dip/issues/new

SCREEN SHOTS

Using DipPy in Solver Studio

Choosing algorithm in Solver Studio

Displaying search tree using GrUMPy

Editing DipPy Model with Eclipse and PyDev

CHANGELOG

Release 0.95.0

• Python 3 support added

Release 0.92.4

• Update dependencies
• Minor bug fixes for DipPy

Release 0.92.3

• Update dependencies
• Minor bug fix
• Add support for Appveyor and Travis
• Install examples with DipPy

Release 0.92.2

• Get rid of pesky global variable DecompInf
• Fix bugs in wedding planner example
• Fix bugs in DipPy to allow returning no solutions, even when an exact subprobem solver is used and to allow no branching candidates when branching.

Release 0.91.6

• Fixed bugs in Wedding Planner example.
• Fixed bug in DipPy having to do with branching.

Release 0.92.1

• Fixed problem with dependency linking

Release 0.91.5

• Fixed bug with SYMPHONY when not all solutions are accepted.

Release 0.92.0

• Substantially re-designed internals
• Renamed classes, functions, and parameters more intuitively
• Eliminated unnecessary "MILP" parameter section and joined it to "DECOMP," as well as making "DECOMP" the default parameter section name.
• Changed parameter setting mechanism to make it possible to pass parameters directly to solvers using native names.
• Added interface to Gurobi
• Added ability to select solver at run-time rather than compile-time.
• In DipPy, the user can now return a status in the subproblem solve to indicate whether the subproblem was solved exactly. Previously, DipPy solved the subproblem to optimality internally whenever no solution was returned, which is unnecessary if the user's subproblem solver is exact. It also means that the user was previously required to provide a full description of the subproblem.

Release 0.91.4

• Fixed bugs in examples
• Updates to dependencies
• Samll bug fixes

Release 0.91.4

• Fixed bugs in examples
• Updates to dependencies
• Samll bug fixes

Release 0.91.3

• Fixes for correctly producing Doxygen documentation

Release 0.91.2

• Fixed issue with master only variable when solving master as an integer program.
• Added ability to generate multiple columns per iteration with SYMPHONY and Cbc.

Release 0.91.1

• Updating dependencies.
• Fix for dependency linking
• Fix to installation with {{{DESTDIR}}}

Release 0.91.0

• Multiple parallel modes added
  • Solution of individual subproblems can be parallelized
  • Multiple subproblems can be solved simultaneously
  • Multiple algorithms can be executed in parallel (branch and cut plus one or more decomposition-based algorithms)
• Warm starting for solutions of subproblems is supported when using SYMPHONY as the subproblem solver.
• Unbounded feasible regions now supported.
• Explicit treatment of master-only variables.

Release 0.9.12:

• Fixed long-standing issues with stand-alone apps

• Small some bug fixes

Release 0.9.11:

• Added some new DipPy examples

• Small bug fixes

Release 0.9.10:

• Fixes to stand-alone app examples

• Fixes to Visual studio files for examples to support property pages

##New Stable Version 0.91:

• Changes to the DipPy callback interface to make it more user friendly

• Changes to the application interface

• Planning for other changes to the internal algorithm

Release 0.9.9:

• Fixes to DipPy build and examples

• Fixes to allow CGL cuts to be generated from within DipPy branch and price.

Release 0.9.8:

• Fixes to DipPy examples

• Support for dependency linking

Release 0.9.7:

• Fixes to DipPy examples

• Support for dependency linking

Release 0.9.6:

• Fixes to allow proper installation of DipPy on Mac OS X

Release 0.9.5:

• Small fixes to DipPy

Release 0.9.4:

• Fixes to parallel subproblem solution mode with OpenMP

Release 0.9.3:

• More updates to build system

Release 0.9.2:

• Updates to build system

Release 0.9.1:

• Fixes to Python installation

• Fix to DipPy

Release 0.9.0:

• DIP now includes DipPy, a Python-based modeling language.

• DIP is now a complete generic MILP solver, capable of automatically detecting block structure and applying a decomposition method.

• DIP and DipPy build out of the box in Windows, OSX, and Linux with a combination of the autotools and a Python setup script.

• There is now support for solving the subproblems in parallel when there is block structure using OpenMP.

• Numerous bug fixes and improvements.

Release 0.82.2:

• Fixes to build system

• Updates to dependencies

Release 0.82.1:

• Fixes to build system

• Updates to dependencies

Release 0.82.0:

• Support for MSVC++ version 10 added.

• Support for BuildTools version 0.7 to incorporate recent enhancements, including proper library versioning in Linux, prohibiting installation of private headers, etc.

• Updated externals to new stable versions of dependent projects.

• Minor bug fixes.

dip's Issues

after round robin loop with no neg rc blocks, should immedaitely update LB as this is valid

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-22 01:21:34

Assignee: @mgalati13

after round robin loop with no neg rc blocks, should immedaitely update LB as this is valid

in retail27, milpblock, if not, performs much worse

Problem with BuildTools configall_system.h

Issue created by migration from Trac.

Original creator: @mosu001

Original creation time: 2010-07-12 00:39:42

From a colleague trying to build Dip with Dippy:

However a new problem came up when compiling, the cause being the Dip/BuildTools/headers/configall_system.h header, that only support microsoft compiler. I did a rather ungraceful fix by deactivating the compiler check, and it seems to work for g++.

Dippy install fails with fatal error: 'Decomp.h' file not found

Building on Mac OS, I am stuck on the python setup.py install step of the install process.

D/c/a/Dip/Dip $ python setup.py install
running install
running bdist_egg
running egg_info
writing requirements to coinor.dippy.egg-info/requires.txt
writing coinor.dippy.egg-info/PKG-INFO
writing namespace_packages to coinor.dippy.egg-info/namespace_packages.txt
writing top-level names to coinor.dippy.egg-info/top_level.txt
writing dependency_links to coinor.dippy.egg-info/dependency_links.txt
reading manifest file 'coinor.dippy.egg-info/SOURCES.txt'
writing manifest file 'coinor.dippy.egg-info/SOURCES.txt'
installing library code to build/bdist.macosx-10.12-x86_64/egg
running install_lib
running build_py
running build_ext
building '_dippy' extension
clang -fno-strict-aliasing -g -O2 -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -D__DECOMP_LP_CLP__ -I/Users/sooheon/Dropbox/code/algos/Dip/include/coin -I/usr/local/gurobi/linux64/include -I/usr/local/cplex/include/ilcplex -I/Users/sooheon/.pyenv/versions/2.7.13/include/python2.7 -c src/dippy/DippyDecompAlgo.cpp -o build/temp.macosx-10.12-x86_64-2.7/src/dippy/DippyDecompAlgo.o
In file included from src/dippy/DippyDecompAlgo.cpp:3:
src/dippy/DippyDecompAlgo.h:4:10: fatal error: 'Decomp.h' file not found
#include "Decomp.h"
         ^
1 error generated.
error: command 'clang' failed with exit status 1

I have tried setting the DYLD_LIBRARY_PATH, and configuring with --enable-static --disable-shared, to no avail.

The COIN_INSTALL_DIR

Which is the install directory? Is it the directory where all the downloaded files were unzipped or is it the build directory that was created for installation?

UtilPrintMemUsage won't compile with gcc 4.5.0 on MSys MinGW

Issue created by migration from Trac.

Original creator: @tkralphs

Original creation time: 2010-05-01 02:22:06

Assignee: @mgalati13

The function UtilPrintMemUsage in UtilMacros.h won't build with gcc 4.5.0 on Msys/MinGW (Windows) and presumably others as well. I commented it out for now, since it's non-essential, but it should be fixed or deleted eventually.

create small simple example to show master-only cols

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-02 15:36:27

Assignee: @mgalati13

create small simple example to show master-only cols

Problem with bounds produced by Alps at node 0

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-08-23 01:39:39

I notice that for a number of problems that I am running ALPS reports the following lower bound at node 0

Alps0240I Processed 0 nodes, has 0 nodes, best relaxed 1e+75, best feasible 16

This is for a minimization problem where all problems have lower bounds of 0 and upper bounds of 1. How can the best relaxed value exceed the best feasible value for a minimization?

Note also the following odd number being reported for the depth

Alps0260I Quality of the best solution found: 16 ; depth 2147483647

Thanks

wishlist - pre/post solve added to DIP

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-17 16:42:02

Version:

I think one of the biggest things missing from DIP – especially for generic MILP is a presolve and postsolve. This is one place you or a student of yours could make a huge contribution. By figuring out a “clean” API for doing IP presolve via CBC? CoinUtils? Cgl? and the corresponding postsolve on a model before handing it to DIP….

You mentioned some student (not Lehigh) that wanted to help out. That would be a major contribution.

Matt

git installing instructions require svn

hello there!

I've just installed dip today. I have on small comment. In the Unix installing instructions, there are svn and git steps.

The git ones go like this:

git clone --branch=stable/0.92 https://github.com/coin-or/Dip Dip-0.92
cd Dip-0.92
git clone --branch=stable/0.8 https://github.com/coin-or-tools/BuildTools/
BuildTools/get.dependencies.sh fetch

First of all, the BuildTools/get.dependencies.sh fetch command requires svn, unless executed with the --git option. I guess it'd make sense to add the option in this command to make it proper git-compatible. So it would end up like: BuildTools/get.dependencies.sh fetch --git.

Secondly, even with the --git option activated, it still tries to use svn at some point. I got the error because I did not have it installed.
I got:

BuildTools/get.dependencies.sh: line 271: svn: command not found

Of course, this is not critical. I just installed the subversion package for Ubuntu and that was it. I just though that maybe it would be nice to make the instructions a little more consistent. Or just take out the git ones altogether.

thanks!

Franco

make fails with -Wpedantic

Hi,

I'm building Dip on Manjaro Linux with 6.3.1 and GNU Make 4.2.1 and after a successful configure step, I get this error when I run make:

../../../../Dip/src/dippy/DippyDecompAlgo.cpp:32:22: error: ISO C++ forbids converting a string constant to 'char*' [-Wpedantic]
pSolutionList);

(there are several more lines like this). I managed to build Dip by commenting the

coin_warn_cflags="-pedantic-errors $coin_warn_cflags"

lines from the configure script, however I don't know if this is supposed to happen.

Set options in C++ code

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-08-10 03:06:54

Add the feature so that the user can set/modify members of the DecompParam class from inside the C++ code.

AP3 example

Issue created by migration from Trac.

Original creator: Uonly

Original creation time: 2010-04-22 04:00:43

Version:

For AP3, it looks like that it is using OLD stuff. It might need some update.

Error: CBC IP solver 2nd status = 7

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-08-09 01:38:33

I am experimenting with Dip and solving a simple plant location problem. In my approach, there is a block (relaxed constraint set) for each plant. There is a relaxed constraint set for each i (plant) and the relaxed constraints for block i are:

x_{ij} \le y_{i} for all j

In this case the y_{i} is a binary variable so the sub-problem is bounded. However I get the error

Error: CBC IP solver 2nd status = 7
COIN Exception [ CBC solver 2nd status ] at ../../../Dip/src/DecompModel.cpp:L330 in DecompAlgoModel::solveOsiAsIp

If I add upper bounds of 1 on the x_{ij} variables it works and runs to completion with the correct optimal solution. However, it seems like I should not have to do this since the constraints x_{ij} \le y_{i} with y_{i} having an upper bound imply this.

Thanks

No bounds checking in MILPBlock

Issue created by migration from Trac.

Original creator: @tkralphs

Original creation time: 2010-04-14 23:10:16

In the MILPBlock example, putting column indices that are too large or too small in the block file causes the code to crash.

gen init vars provides a valid LB, use it

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-06 13:17:28

Assignee: @mgalati13

GenInitVars provides a valid LB (sum the LBs of each block) which in some cases, DW might take a long time to converge to. Use this as the starting LB - sometimes it is very good.

enhance solveRelaxed to allow parallel processing of blocks

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-07-21 04:25:30

Currently solveRelaxed() is processed sequentially for each block.

Is it possible to have a method something like:

solveRelaxed( const double *costCoeff, DecompVarList & newVars)

where costCoeff is all of the cost coefficients across blocks and
likewise for newVars. Then it is up to the user to find the solution to the blocks either sequentially or in parallel? The user should still be able to pass back an extreme point for each block so there is a convexity row in the master for each block.

Thank you,

make explicit examples for different variants of GAP example for illustration

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-02 14:24:00

Assignee: @mgalati13

CC: [email protected]

create separate simple example for each case:
(1) using pisinger
(2) using built-in milp
(3) dense blocks
(4) sparse blocks

Typo DecompMain

Hi,
I am afraid there is a typo in DecompMain.cpp, lines 583 to 585, stopping the build in MSVC.

"SYMPONY" --> SYMPHONY

ifdef DIP_HAS_CLP --> #ifdef DIP_HAS_SYM

By the way, the nice work!
Eduardo

Clean up examples

Issue created by migration from Trac.

Original creator: @tkralphs

Original creation time: 2010-04-14 22:47:27

Assignee: somebody

Version:

A number of the examples don't have READMEs and I don't think all of them work out of the box as the README indicates. For the READMEs that exist, there are some cut and paste errors that confuse matters. We should go through and just check each one and clean them all up.

use row names or ids for block id file

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-14 23:36:35

Assignee: @mgalati13

use row names or ids for block id file

VRP/TSP concorde - linked LP solver should be driven by DIP config, or example config

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-06 17:11:37

VRP/TSP concorde - linked LP solver should be driven by DIP config, or example config.

Currently, the user must config concorde manually.

In case of VRP, it uses a specific lib of qsopt. This won't be portable.

new option to use col pool for master as IP

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-28 19:56:52

CC: [email protected]

Suggested by Rob Pratt.

After compress columns and put in col-pool, when solving master as IP, reintroduce back the columns in pool (node feasible ones). Might help find integer feasible solution.

doPriceCut gives infeasible when doCut and doRelaxCut solve

Issue created by migration from Trac.

Original creator: @mosu001

Original creation time: 2010-05-29 05:59:37

Version: 0.8.7

When using Dippy to generate a problem, it will solve if using doCut or doRelaxCut or default option, but gives infeasible when using doPriceCut. I have attached the LP file generated by Dippy.

user provided init var in block case - force incubment into system

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-18 13:40:38

Assignee: @mgalati13

Current design - if user provides init var in full space and we have blocks, we break it out into partial blocks and rely on solutionUpdateAsIP to recombine it to get first globalUB. Should force this into system before that so we don't have to rely on that to get that UB back.

need sparse version of milpblock block creation (see Alper's SSR)

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-27 21:03:13

Assignee: @mgalati13

need sparse version of milpblock block creation (see Alper's SSR)

redo branching scheme for PC

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-14 23:37:11

Assignee: @mgalati13

Dynamically create branching rows.

RoundRobinStategy is not implemented

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-22 01:19:27

RoundRobinStategy is not implemented

Problem with 0 integer variables

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-08-01 03:48:50

Hi:

I understand Dip is an integer programming package, but it seems like it should still work if I have a block angular LP and want to apply Dantzig Wolfe Decomposition. When I try to solve a problem (for example the SmallIP test problem or the GAP test problem) and do not declare integer variables I get the error

COIN Exception [ modelCore->integerVars.size() > 0 ] at ../../../Dip/src/DecompBranch.cpp:L38 in ::

Thanks

UseSparse and InitVarsWithIP not working together

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-06-02 00:03:37

Assignee: @mgalati13

UseSparse and InitVarsWithIP not working together

Call to setModelRelax results in duplication of variables in activeColumns

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-08-09 02:16:00

A call to setModelRelax results in a duplication of the variables in the vector of activeColumns.

To see this add the line of code

 UtilPrintVector((*mdi).second->activeColumns, m_osLog);

after the call

  setModelRelax((*mdi).second,
                "relax" + UtilIntToStr((*mdi).first),
                (*mdi).first);

in Dip/examples/MILPBlocks/MILPBlock_DecompApp.cpp. See, for example, the list below generated for one block:

85 171 257 343 429 515 601 687 773 859 945 1031 1117 1203 1289 1375 1461 1547 1642 1728 1814 1900 1986 2072 2158 2244 2330 2416 2502 2588 2674 2760 2846 2932 3018 3104 3190 3276 3362 3448 3534 3620 3706 3792 3878 3964 4050 4136

85 171 257 343 429 515 601 687 773 859 945 1031 1117 1203 1289 1375 1461 1547 1642 1728 1814 1900 1986 2072 2158 2244 2330 2416 2502 2588 2674 2760 2846 2932 3018 3104 3190 3276 3362 3448 3534 3620 3706 3792 3878 3964 4050 4136

make api for master-only vars more simple

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-02 02:26:53

CC: [email protected]

Good question... It's actually a bit messy to deal with this because of the way the framework is designed.

I don't have any explicit (simple) examples of this and should probably add one. For now, you can look at MILPBlock which deals with this case. I call it "master-only vars".

Look at MILPBlock_DecompApp::createModelMasterOnlys2( ) as an example of how to deal with this. Essentially, I just need the LB and UB of the vars. But since the design is driven off the DecompConstraintSet object, you need to declare an empty object for each var. I should probably make this easier for the user by just asking for a list of them. I’ll add a ticket to fix that in future.

for(vit = masterOnlyCols.begin(); vit != masterOnlyCols.end(); vit++){
i = *vit;

  //THINK:
  //  what-if master-only var is integer and bound is not at integer

  DecompConstraintSet * model = new DecompConstraintSet();
  model->m_masterOnly      = true;
  model->m_masterOnlyIndex = i;
  model->m_masterOnlyLB    = colLB[i];
  model->m_masterOnlyUB    = colUB[i];
  //0=cont, 1=integer
  model->m_masterOnlyIsInt = integerVars[i] ? true : false;
  if(m_appParam.ColumnUB <  1.0e15)
     if(colUB[i] >  1.0e15)
        model->m_masterOnlyUB = m_appParam.ColumnUB;
  if(m_appParam.ColumnLB > -1.0e15)
     if(colLB[i] < -1.0e15)
        model->m_masterOnlyLB = m_appParam.ColumnLB;

  m_modelR.insert(make_pair(nBlocks, model));
  setModelRelax(model,
                "master_only" + UtilIntToStr(i), nBlocks);
  nBlocks++;

}

-----Original Message-----
From: Kipp Martin [mailto:[e-mail]]
Sent: Saturday, July 31, 2010 3:20 AM
To: Matthew Galati
Subject: Re: Dip Questions

Hi Matt:

In my application I have a set variables in the core constraints that do not
appear in any of the blocks. So something like this

min cx

s.t

A'x >= b1
A''x + By >= b2

I assume that this will not mess things up, correct?

Thanks

boost for VRP/TSP - download only those files needed rather than all of boost

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-06 17:10:14

boost for VRP/TSP - download only those files needed rather than all of boost

Artificial variables in createMasterProblem cause crash in isParallel

Issue created by migration from Trac.

Original creator: @mosu001

Original creation time: 2010-05-26 01:46:06

Version:

This is in version 0.8.6.

Artificial variables created at the end of the DW formulation in DecompAlgo::createMasterProblem in for loop starting at line 820 have a zero length column when being checked in DecompVarPool::isParallel. This causes a crash in while loop starting on line 98.

Recommend either changing while condition to

index2 < len2 && index1 < len1

or (if not too many comparisons wanted) take break statement

     if(index2 >= len2 || index1 >= len1)
        break;

out of if (j1==j2) block and placing directly after while(1)

Allow use of SYMPHONY as MIP solver

Issue created by migration from Trac.

Original creator: @tkralphs

Original creation time: 2010-08-20 18:23:41

Hook up SYMPHONY through OSI so that it can be used as the subproblem solver.

integrate finder for block angular to MilpBlock

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-06 15:02:10

Version:

integrate finder for block angular to MilpBlock

masterMatrixAddArtCol is too slow for large scale

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-05 00:50:28

Assignee: @mgalati13

switch to appendCols from repeated calls to appendCol for dramatic speed-up

create SmallIP example to illustrate multiple blocks (like MILPBlock)

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-09 13:07:51

Assignee: @mgalati13

create SmallIP example to illustrate multiple blocks (like MILPBlock)

make does work on Mac OS X

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-07-19 05:49:56

Hi:

I am trying to build Dip on Mac OS X 10.6.1 using gcc 4.2.1. I was able to run configure with no problem. However make results in the error below. I can build numerous other COIN-OR project on this platform that also refer to malloc.h but those reference are in ifdef. Below is the error:

In file included from ../../../Dip/src/DecompParam.h:21,
from ../../../Dip/src/DecompApp.h:22,
from ../../../Dip/src/DecompAlgo.cpp:17:
../../../Dip/src/UtilMacros.h:41:20: error: malloc.h: No such file or directory
../../../Dip/src/DecompAlgo.cpp: In member function 'virtual void DecompAlgo::createMasterProblem(DecompVarList&)':
../../../Dip/src/DecompAlgo.cpp:712: warning: unused variable 'c'
../../../Dip/src/DecompAlgo.cpp: In member function 'std::vector<double*, std::allocator<double*> > DecompAlgo::getDualRays(int)':
../../../Dip/src/DecompAlgo.cpp:4016: warning: unused variable 'ray'
../../../Dip/src/DecompAlgo.cpp: In member function 'virtual void DecompAlgo::addVarsFromPool()':
../../../Dip/src/DecompAlgo.cpp:5464: warning: unused variable 'n_colsAfter'
../../../Dip/src/DecompAlgo.cpp: In member function 'DecompStatus DecompAlgo::solveRelaxed(const double*, const double*, double, int, bool, DecompAlgoModel&, DecompSolverResult*, std::list<DecompVar*, std::allocator<DecompVar*> >&)':
../../../Dip/src/DecompAlgo.cpp:6347: warning: unused variable 'isRelaxFeas'
make[2]: *** [DecompAlgo.lo] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all-recursive] Error 1

Thanks

column generator use multiple solutions if MILP solver provides it

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-04 00:43:31

Assignee: @mgalati13

column generator use multiple solutions if MILP solver provides it, just check that neg RC

option to not add cols that are close to parallel

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-11 12:20:54

Assignee: @mgalati13

option to not add cols that are close to parallel

Negative bounds causing problem

Issue created by migration from Trac.

Original creator: @mosu001

Original creation time: 2010-05-26 11:48:39

Version: 0.8.6

I have a problem that at this stage is only a master problem with no subproblems. However, some of the variables have negative lower bounds. This causes a problem in solutionUpdate in DecompAlgo.cpp as any primal column with a negative value causes

(*m_osLog) << "ERROR: NEGATIVE LAMBDA, but Osi returns as optimal"
<< " assume it was meant to be infeasible." << endl;

I can change my formulation to only have non-negative bounds, but perhaps a warning might be good?

COIN_HAS_CBC x DIP_HAS_CBC

Hi,
My suggestion is change preprocessor names COIN_HAS_XXX (XXX = CBC, SYMPHONY, ...) by DIP_HAS_XXX in decompModel.cpp, as COIN_HAS_CBC is already defined in other lib, and DIP has its own solvers definitions (DIP_HAS_XXX).

Thanks,
Eduardo

problem with MILPBlock example on Mac OS X.

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-07-20 02:11:38

When I run

./decomp_milpblock --param milpblock.parm

I get the error below. I was however able to run MILPBlock after doing a new configure with

--enable-static --disable-shared

so it works with static libs and not dynamic.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

dyld: Symbol not found: __ZN12CbcHeuristic13shouldHeurRunEi
Referenced from: /Users/kmartin/coin/dip-trunk/vpath/lib/libCbcSolver.0.dylib
Expected in: flat namespace
in /Users/kmartin/coin/dip-trunk/vpath/lib/libCbcSolver.0.dylib
Trace/BPT trap

user should not have to create empty matrix for case where oracle is defined

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-08-02 02:51:16

Assignee: @mgalati13

CC: [email protected]

Ok. The user should not even have to do that (create empty matrix).

----- Original Message -----
From: Kipp Martin <[e-mail]>
To: Ted Ralphs <[e-mail]>; Matthew Galati
Sent: Thu Jul 29 04:12:00 2010
Subject: Dip Follow Up

Hi Matt/Ted:

I think I figured out my answer to question 1 in the previous email.
In the GAP example, I just commented out the lines of code where the knapsack coefficients are explicitly generated. I guess just having

model->M = new CoinPackedMatrix(false, 0.0, 0.0);
CoinAssertHint(model->M, "Error: Out of Memory");
model->M->setDimensions(0, nCols);

does the trick, but it is not clear from the documentation what I should be doing.

cutgen OSI object is currently adding in A' blocks by default

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-05 01:14:01

Assignee: @mgalati13

unless isSparse - this might be overkill and is probably a really bad idea for memory

add user API to allow user to select which block to process next

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-04-21 22:19:16

Assignee: @mgalati13

add user API to allow user to select which block to process next

Setting a time limit with dippy

Hi there!

First of all, congratulations on the project. I've been playing for a couple of days and it is very nice!

I am using dippy to solve a vehicle routing problem and it would be nice if I could set a time limit for the solver via dippy (i.e. in my python code) since it is a hard problem to solve as the number of clients increases.
Where can I set this parameter or, if there is no parameter, could you give some hints on how I can make it happen?
I tried to check a portion of the codes inside the dippy folder, but I could not find any clues for this question.
I apologize if this question it is not supposed to be done here.

Best regards!

resdesign LB calculation to use LBs of subproblems, not varRedCost

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-04 01:32:05

Assignee: @mgalati13

resdesign LB calculation to use LBs of subproblems, not varRedCost - this will allow stopping on gap in subproblems and updating LB when subproblems are not solved exactly

Problem building examples/GAP

Issue created by migration from Trac.

Original creator: kmartin

Original creation time: 2010-07-20 09:21:32

I followed the instructions for building GAP and on Mac OS X 10.6 I get the error below when executing make. (PS -- I also got the malloc.h error reported in a previous ticket but just copied a malloc.h file into

ExternalSolvers/Knapsack/Pisinger

to fix the problem. )

%%%%%%%%%%%%%%%%%

'`../../../../Dip/examples/GAP/GAP_Main.cpp
../../../../Dip/examples/GAP/GAP_Main.cpp: In function ‘int main(int, char**)’:
../../../../Dip/examples/GAP/GAP_Main.cpp:84: error: invalid conversion from ‘int’ to ‘const DecompSolution*’
../../../../Dip/examples/GAP/GAP_Main.cpp:84: error: initializing argument 1 of ‘virtual DecompSolverResult* DecompAlgo::solveDirect(const DecompSolution*)’

some bug in DualStab implmentation

Issue created by migration from Trac.

Original creator: @mgalati13

Original creation time: 2010-05-06 16:44:23

Assignee: @mgalati13

MILPBlock, atm_20_100_1.tight with CPX12, defaults optimal=2,463,554. With DualStab=1, DualStabAlpha=0.3, optimal=2,463,699.