Pass constraint names to CPLEX

[DillonJ]

Hi,

All CPLEX solver messages are reported with generic constraint names e.g. c23474

It makes it very difficult to diagnose infeasibilities and such.

Is there a plan to pass the JuMP constraint names on to CPLEX? It seems possible as in GAMS for example I get more helpful names.

[mlubin]

Yes, these names should be passed to CPLEX.

[chengg04]

This should be possible as CPLEX API has a function called SetName() for such a purpose. We can have a wrapper for this.

[DillonJ]

Hi folks,

Thanks @odow and @mlubin for looking at this.

I'm wondering this is working properly or if I'm not using it correctly?

I have named all the constraints and variables in my model and I can see the names when I write an MPS file using write_to_file(). However, CPLEX still reports diagnostic information using generic names as follows:
Integer infeasible column 'x6467'.

Is there something specific I need to do ? Is it possible to relate the CPLEX name above to the MPS names in any way?

[odow]

We pass variables names to CPLEX if they are ASCII: https://github.com/JuliaOpt/CPLEX.jl/blob/5ae4628446470fa0a46438cdfe577155dbcfd54c/src/MOI/MOI_wrapper.jl#L599-L603

So either your variable name is non-ascii, or it has an empty name (i.e., "").

If you can provide a minimal working example (using 2 or 3 variables, not 6000), open a new issue.

[DillonJ]

Ahhh...

julia> isascii("unit_flow_capacity(CLN,_ELEC_1911_CUI_110,_to_node,_2000-01-01T00:00:00_�_2000-01-01T01:00:00)")
false

Maybe some sanitisation to be done...

[DillonJ]

So that didn't solve it.

Here is a minimum example :

using JuMP
using CPLEX
m = Model(CPLEX.Optimizer)
@variable(m, bin_var1, Bin)
@variable(m, bin_var2, Bin)
con1=@constraint(m, constraint1, bin_var1 + bin_var2 >= 2)
con2=@constraint(m, constraint2, bin_var1            == 0)
@objective(m, Max, 5bin_var1 + 3bin_var2)
optimize!(m)
write_to_file(m, "cplex_min_example.mps")

Which outputs

Warning:  Non-integral bounds for integer variables rounded.
Warning:  No solution found from 1 MIP starts.
Retaining values of one MIP start for possible repair.
Warning:  Non-integral bounds for integer variables rounded.
Row 'c2' implies binary variable 'x2' is greater than 1.
Presolve time = 0.00 sec. (0.00 ticks)

And in the MPS file:

NAME          
ROWS
 N  OBJ
 G  constraint1
 E  constraint2
COLUMNS
    MARKER    'MARKER'                 'INTORG'
     bin_var2 constraint1 1
     bin_var2 OBJ      -3
    MARKER    'MARKER'                 'INTEND'
    MARKER    'MARKER'                 'INTORG'
     bin_var1 constraint1 1
     bin_var1 constraint2 1
     bin_var1 OBJ      -5
    MARKER    'MARKER'                 'INTEND'
RHS
    rhs       constraint1  2
    rhs       constraint2  0
RANGES
BOUNDS
 BV bounds    bin_var1
 BV bounds    bin_var2
ENDATA

Even if I explicitly use the set_name() function, CPLEX still reports anonymous constraint and variable names.

[odow]

What happens if you use m = direct_model(CPLEX.Optimizer())

[DillonJ]

Ah, interesting :

Warning:  Non-integral bounds for integer variables rounded.
Warning:  No solution found from 1 MIP starts.
Retaining values of one MIP start for possible repair.
Warning:  Non-integral bounds for integer variables rounded.
Infeasibility row 'constraint1':  0 >= 1.
Presolve time = 0.00 sec. (0.00 ticks)

[odow]

@blegat, isn't this sufficient to have JuMP copy names to the optimizer?

https://github.com/JuliaOpt/CPLEX.jl/blob/5ae4628446470fa0a46438cdfe577155dbcfd54c/src/MOI/MOI_wrapper.jl#L341-L345

[blegat]

The names are not passed to the optimizer, they are only stored in the cache. It's not related to CPLEX