PuLP does not correctly read a class of MPS files in MILIB2017

[jacurick19]

Details for the issue

What did you do?

I used PuLP to read the MPS files for MIPLIB2017 problems such as beasleyC3. I then used PuLP to call solvers to optimize the model. I also used PuLP to write the MPS file back to disk without optimizing it and used multiple MIP solvers to solve the MPS file that was written back to disk.

What did you expect to see?

I expected to see the objective value of 754, which is known to be optimal. If I use a solver on the MPS file downloaded directly from MIPLIB2017, this value is achieved.

What did you see instead?

If PuLP is used to dispatch a solver after reading the MPS file, the solution reported is less than 50. If a MIP solver is used on its own to solve the MPS file that PuLP wrote to disk after reading in the beasleyC3 problem's MPS file, the MIP solver reports a solution that is less than 50.

This implies that PuLP is not reading the file correctly. I found that this is the case for multiple MIPLIB2017 problems that have no integer variables but do have binary variables.

Useful extra information

The info below often helps, please fill it out if you're able to. :)

What operating system are you using?

• [ x] Windows: (11)

I installed PuLP via:

• [ x] pypi (python -m pip install pulp)

Did you also

• [x ] Searched for an existing similar issue: https://github.com/coin-or/pulp/issues?utf8=%E2%9C%93&q=is%3Aissue%20

[pchtsp]

hello, thanks for reporting.
Did you compare the original mps file to the one exported by pulp?
Is the solution given by pulp different or just the objective function value?

[jacurick19]

Hello, thank you for quickly responding. I did compare these files, and while I did see some differences, I didn't see anything that obviously would be causing the issue.

The solution solvers produce when solving the PuLP-read MPS file seems to be the solution to the relaxation of the problem, but when solvers read the MPS directly they correctly solve the integer version.