Dev: 
This is a small tool that converts scalar optimization model with different formats into a JuMP model script (.jl). The scalar model is unlike a modeling script that carries the model information with sets, looped-constraints, and complex index variables. You will find this tools useful when benchmarking your algorithm using established instance libraries. Please see some of the converted MINLP instances in MINLPLibJuMP.jl.
This is a developing repo as more formats and exceptions are considered. The mission is to provide
correct and flexible conversion to JuliaOpt users.
Pkg.clone("https://github.com/jac0320/toJuMP.jl.git")
gms2jump("*.gms")
mod2jump("*.mod")
The default output path is Pkg.dir()/toJuMP/.jls/. Multiple options are provided for conversions. The default output path is Pkg.dir()/toJuMP/.jls/.For example,
gms2jump("*.gms", mode="index", ending="m=m", quadNL=true, outdir="", loopifpossible=true)
-
mode: ("raw"or"index") index method for variables. If"raw", then variables will not be indexed and original script variable string ("x15","i26") is defined. If"index", then index will be used to construct variables such as"x[15]"or"i[26]". -
ending: control of the ending line of generated scrip. By default, this is"m=m", which means when you include the generated script, the JuMP model will be returned. -
quadNL: build quadratic model using@NLconstraint. By default (JuMP 0.18-),@constraintcan handle quadratic constraints by automatically expand the expression. User can choose to keep the original expression through@NLconstraint, which will yields a nonlinear JuMP model type. -
outdir: user-defined output dir for.jlfiles. No need to give the problem name. -
loopifpossible: write variables attributes (bounds, warm-start values using loops if possible).
-
Supports for dual warm start is temporarily disabled (will consider with JuMP 0.19)
-
Semi-continous variables is not supported
-
Unsupported operators:
arctan,ceil,errorof,floor,mapval,max,min,mod,normal,round,sign,trunc,uniform. These variables are disabled since they can be achieved through various modeling techniques. Currently,toJuMP.jldoes not assume/apply these modeling techniques by default. -
Variable attributes such as
scale,prior,stageis disabled given these are solver-dependent attributes.