Variable Declaration and Index Sets

[dirkdegel]

Hi Dirk,

I'm curious about the way variables are declared in rmpl compared to ompr and why the way how to do it changed or what the idea is behind this change.

rmpk

x <- model$add_variable(i = 1:n, j = 1:n, k = 1:9, type = "binary")

vs ompr

add_variable(model, x[i, j, k], i = 1:n, j = 1:n, k = 1:9, type = "binary")

I personally find the "ompr" way much better readable and much more consistent with regards to how the objective and the constraints are set and added.

Also, from reading the Julia/JuMP documentation: Is there not the risk of "mixing" R variables with "rmpk variables", e.g. x should be a R variable whereas in add_variable(model, x) x is something like a reference from the model variable x to the variable representation in the solver (I hope that makes sense). I think it would be beneficial if the syntax would be consistent and look something like

add_variable(model, x[i], I = index_set, name = "...")
add_constraint(model, expression, quantifier, name = "...")
add_objective(model, ...) or set_objective()

This should be equivalent to

model$add_variable(x[i], I = index_set, name = "...")
model$add_constraint(expression, quantifier, name = "...")
model$add_objective(, ...)

Thanks a lot for your great work!
Dirk

[dirkschumacher]

Thanks for the feedback. #33 and PR #36 are related to this interface idea.

I also like the ompr interface, but I would also like to have direct control over the model/ without any non-standard evaluation. One problem with e.g. add_variable(model, x[i, j, k], ...) is that it if used within a package you will get an R CMD check error, due to unassigned variable usage.

But yeah, I want to offer an api that is very similiar (or even fully compatible) to ompr but also have a more lower level interface using the R6 class. At least for now. Some steps have been made in PR #36 :)