Forum Discussion
Optimizer not loading the entire MPS file
Hi all,
I'm playing around with a large MIP problem, and trying to compare Xpress and Cplex (2018 versions) in terms of performance for future work.
MIP problem is formulated in GAMS, and solved directly without any issue using Cplex. As the Xpress licence is on another computer, I export a MPS file with GAMS and I'm trying to read and solve it with Xpress Optimizer.
For small version of the MIP problem, everything went well. For the full problem, when trying to load the MPS with Optimizer, I get unusual warnings and errors:
Warning: Ignoring repeated specification for col : "x1000000" in line ... while reading in free format the file 'MIP.mps'
many times, and then,
Problem Statisticts
- 4793730 (0 spare) rows
- 9999999 (994529 spare) structural columns
- 38151669 (58691964) non-zero elements
Error: Bound conflict for colum xXXXXX : 69.5 > 42
Error : Input aborted
I'm suprised the problem raises an error. It has been solved to optimality on Cplex, and smaller versions of it did not cause any trouble. I would expect Cplex to also raise an error if the problem was wrongly formulated.
I'm also surprised that by chance, it is variable "x1000000" causing a problem and that only 9999999 structural columns where found. I don't believe in coincidence! The beginning of the MPS file reads:
* MIP written by GAMS Convert at 06/01/24 10:57:30
*
* Equation counts
* Total E G L N X C B
* 4793729 962465 3064671 766593 0 0 0 0
*
* Variable counts
* x b i s1s s2s sc si
* Total cont binary integer sos1 sos2 scont sint
* 19945289 19944947 342 0 0 0 0 0
* FX1977036 1976998 38 0 0 0 0 0
*
* Nonzero counts
* Total const NL DLL
* 96843632 96843632 0 0
*
NAME Convert
*
* original model was minimizing
So, I should have more columns, right? What is happening here? Could it be that Optimizer has a limit on the number of columns/variables it can read? Is it related to a control parameter, to the MPS file or to a limited licence maybe?
Any help or suggestion would be greatly appreciated!
Milien
4 Replies
Did you try reading this MPS file CPLEX as well? It would be interesting to see whether CPLEX gives the same result when solving directly and/or when reading from the MPS file.
The problem you see with Xpress could be a problem with the MPS file or with Xpress itself. Could you share the (compressed) MPS file?
Hello Daniel,
Thank you for your help. I'll see if I can solve the MPS file with CPLEX.
In the meantime, here is the compressed MPS file generated by GAMS.
At first glance the file looks at least strange. For example you have these lines (line numbers to the left of the colon):
42945485: x1000000 e192274 -0.3953 42945486: x1000000 e63 -77.16725385 42945487: x1000000 e104258 1 42945488: x1000000 e192274 -0.3953 42945489: x1000000 e63 -77.16725385 42945490: x1000000 e104259 1 42945491: x1000000 e192274 -0.3953 110922508: UP bnd x1000000 182 110922509: UP bnd x1000000 182 110922510: UP bnd x1000000 182 110922511: UP bnd x1000000 182 110922512: UP bnd x1000000 182 110922513: UP bnd x1000000 182So either this is highly repetitive or the column names got truncated to 8 characters.
Moreover, Xpress issues this warning
?18 Error: Bound conflict for column x1573539 : 69.5917 > 42And indeed the file says
102781311: FX bnd x1573539 69.591674194 116580618: UP bnd x1573539 268.02631579 116580619: UP bnd x1573539 268.02631579 116580620: UP bnd x1573539 268.02631579 116580621: UP bnd x1573539 268.02631579 116580622: UP bnd x1573539 268.02631579 116580623: UP bnd x1573539 268.02631579 116580624: UP bnd x1573539 268.02631579 116580625: UP bnd x1573539 42 116580626: UP bnd x1573539 42 116580627: UP bnd x1573539 42In other words, many different bounds are specified for x1573539 in the MPS file. The first line fixes the variable to 69.591674194. Then a bunch of lines sets upper bound to 268.02631579 and then another bunch of lines sets the upper bound to 42.
This smells a lot like a problem with the MPS export of GAMS. Is there maybe an option you have to give when you have a lot of variables so that names do not get truncated to 8 characters? Could you check with GAMS support people? The MPS file definitely looks unexpected.
Thank you very much, Daniel, for taking the time to delve into the MPS file. I admit I only took a brief look at the bound conflict, and since MPS is a scalar format, it certainly doesn't help in understanding why two bounds were applied to the same variable.
You identified the exact problem. I used a FixedMPS format in GAMS, which I believe is quite old and does not allow more than 8 characters for variables, with fields starting in fixed columns (2, 5, 15, 25, 40, and 50). Therefore, the variable x10000000 became x1000000, leading to conflicts and an incorrect formulation. This was not an issue for smaller versions of my problem since they did not reach 10,000,000 variables. The correct parameter to use is CplexMPS in GAMS, to generate a free (and not fixed) MPS file.
It is working perfectly now, and it also generates files that are 40% smaller. Again, thank you very much, Daniel. Your help was invaluable!
