Write .NL file containing JuMP model

Specific Domains Optimization (Mathematical)
,
1

Dear all,

I would ask you if there exists an easy way to write the .nl file corresponding to a JuMP problem. I already checked the solution proposed in https://discourse.julialang.org/t/how-to-write-nl-file-containing-jump-problem-using-amplnlwriter-jl/2954, but it seems no more working.

Thanks very much. Best regards,
Luca

2 Likes
2

Try write_to_file(model, "model.nl")

3

This isnt working for me, no matter which Optimizer I use.

Writing to .lp or .mps files is fine, but those are rejected by AMPL solvers. Where writing to .nl files results in weird formatting (that causes parse errors when I try to feed it to a solver.)

This is what my .nl file looks like, when I generate it with JuMP.write_to_file(model, "pentomino.nl")

g3 1 1 0
 7865 9126 1 0 12 0
 0 1
 0 0
 0 0 0
 0 0 0 1
 7865 0 0 0 0
 18919288 7865
 0 0
 0 0 0 0 0
C0
n0
C1
4

Hi @ink,

What’s the error? And with what solver? What happens if you use AmplNLWriter.jl?

Do you have a reproducible example? Can you send me the pentomino.nl file that gets written? Your Jacobian looks very dense.

5

I use AmplNLWriter with Bonmin_jll and it printed the same .nl file as when I just use HiGHS.Optimizer (without AmplNLWriter). I exported the model “normally” with JuMP.write_to_file(model, nl_path) - but when trying to load this file into AMPL in python (which supposedly accepts .nl files) with ampl.read(nl_filename), it threw a parse error. (The .nl file looks like it’s missing spaces or something, I’m not sure, I’ve never used this format before.)

I’ll try and prepare a minimal reproducible example, sure, since my codes are like 300+ lines long atm. Thank you for your interest in helping !

It’s also possible that ampl simply doesn’t accept .nl files at all? I know it prefers its own .mod and .dat files, but those are not trivial to generate with JuMP - I have to generate them manually from Julia by printing strings, which is kind of annoying (and error prone).

Edit: worth mentioning, my problem is linear. So I want to use specialized NL solvers like CPLEX and stuff, not Bonmin and those ones.

6

I haven’t used AMPLs Python API.

There might be a bug in how we’re writing the nl file or you might be calling AMPL wrong. Contact their support?

Why go via nl? Why not CPLEX.jl?

1 Like
7

Thanks for the help in DMs, seems that AMPL’s Python API doesn’t support the .nl file format at all, that’s where the problem lies.

To anwser why not CPLEX directly: I have an AMPL license which provides online CPLEX solving, but don’t have a CPLEX license personally, and my model is much too big for the free version (1000x1000 max), that’s why I used AMPL - and I was told it would read .nl files, not just .mod and .dat files, which was my mistake! Problem closed :grinning_face_with_smiling_eyes: