Playground to experiment with convenience functions for working efficiently with sparse variables in JuMP.
For large numbers of dimensions and indices, problem construction is much more efficient when creating just the necessary variables:
using SparseHelper, JuMP
ts = [(i=:A,j=2,k=5),(i=:B,j=5,k=4),(i=:C,j=3,k=5)]
I,J,K = sparsehelper(ts,3)
m = Model()
@variable(m, x[i=I,j=J[i],k=K[i,j]])
WIP: Macro to create sparse variables directly:
m = Model()
@sparsevariable(m,x[i,j,k] for (i,j,k) in ts)Alternatively, one may wish to create the variables over the tuples directly:
m = Model()
@variable(m, x[ts])That makes working with indices later a bit more involved, though. This can be made more convenient with a few helper functions:
# With named tuples.
# Select tuples where k=5:
select(ts,(;k=5))If you don't use named tuples, use :* as wildcard:
# If not using named tuples
nnts = values.(ts)
select(nnts,(:*,:*,5))WIP: For convenience, sum over tuples that match the search criterion:
@constraint(m,sum_where(x,(;k=5)) <=10)
@variable(m,y[ts])
@constraint(m,sum_where(y,(:*,:*,5)) <=10)Feedback and suggestions for improvements is welcome!