CBC solver regression: the exact same model solved on 7.0.6546 is proven infeasible in 7.5.7466

[ihadanny]

What version of OR-tools and what language are you using?
Version: 7.5.7466
Language: Python

Which solver are you using (e.g. CP-SAT, Routing Solver, GLOP, BOP, Gurobi)
CBC_MIXED_INTEGER_PROGRAMMING

What operating system (Linux, Windows, ...) and version?
mac, Darwin Kernel Version 18.7.0

What did you do?

I'm creating and solving a model with 7.0.6546.
I reach a solution after several seconds best objective 4110.8937 (best possible 3724.4987).
I saved it as MPS format (attached).
I then upgrade my version to 7.5.7466.
I execute the exact same code and model and get Result - Problem proven infeasible.

What did you expect to see
the same

bad_7.5.7466.mps.txt
good_7.0.6546.mps.txt

[lperron]

Are you sure it was feasible in the first place ? Can you reproduce using the cbc binaries ? Laurent Perron | Operations Research | lperron@google.com | (33) 1 42 68 53 00 Le jeu. 12 mars 2020 à 07:41, Ido Hadanny <notifications@github.com> a écrit :

…

*What version of OR-tools and what language are you using?* Version: 7.5.7466 Language: Python *Which solver are you using (e.g. CP-SAT, Routing Solver, GLOP, BOP, Gurobi)* CBC_MIXED_INTEGER_PROGRAMMING *What operating system (Linux, Windows, ...) and version?* mac, Darwin Kernel Version 18.7.0 *What did you do?* I'm creating and solving a model with 7.0.6546. I reach a solution after several seconds best objective 4110.8937 (best possible 3724.4987). I saved it as MPS format (attached). I then upgrade my version to 7.5.7466. I execute the same code and get Result - Problem proven infeasible. *What did you expect to see* the same bad_7.5.7466.mps.txt <https://github.com/google/or-tools/files/4322654/bad_7.5.7466.mps.txt> good_7.0.6546.mps.txt <https://github.com/google/or-tools/files/4322657/good_7.0.6546.mps.txt> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1919>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3IVMEHBWXP3R324OD3RHB72XANCNFSM4LGFNBZA> .

[ihadanny]

what do you mean by

reproduce with the CBC binaries?

is there some command line I can use to read and solve a MPS file?
because my application pre-processing is pretty complex and it's currently not-separable from the modeling API calls.

[lperron]

https://bintray.com/coin-or/download/Cbc/#files Laurent Perron | Operations Research | lperron@google.com | (33) 1 42 68 53 00 Le jeu. 12 mars 2020 à 09:21, Ido Hadanny <notifications@github.com> a écrit :

…

what do you mean by reproduce with the CBC binaries? is there some command line I can use to read and solve a MPS file? because my application pre-processing is pretty complex and it's currently not-separable from the modeling API calls. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1919 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3PKD4FTS3B4RVOMJODRHCLR7ANCNFSM4LGFNBZA> .

[ihadanny]

using the cbc command line I was able to solve both the bad file (produced by 7.5) and the good file (produced by 7.0)

[Mizux]

which version of cbc did you use ?
they currently have a regression with the last stable version IIRC
EDIT: see: coin-or/Cbc#286
And we bump it in v7.1 https://github.com/google/or-tools/releases/tag/v7.1

[ihadanny]

very strange. I downloaded 7.5 from https://github.com/google/or-tools/releases/download/v7.5/or-tools_MacOsX-10.15.2_v7.5.7466.tar.gz, and executed:

make build SOURCE=examples/cpp/solve.cc
bin/solve --solver cbc --input bad_7.5.7466.mps --time_limit_ms 5000

and got:

Status      : MPSOLVER_FEASIBLE
Objective   : 4.848683892999999e+03
BestBound   : 3.724498655000000e+03

[ihadanny]

so it works from the C++ API.
Is there some way I can do the same from the python API?

[cloudmu]

which version of cbc did you use ?
they currently have a regression with the last stable version IIRC
EDIT: see: coin-or/Cbc#286
And we bump it in v7.1 https://github.com/google/or-tools/releases/tag/v7.1

If the root cause is the CBC regression, this issue would affect not only Python, but also or-tools API in other languages (Java and C#).

[cloudmu]

@Mizux is this bug resolved? Thanks.