README.md
In dirkschumacher/rmpk: Model Linear and Quadratic Programs

MIP Modelling in R

rmpk is a lightweight package to model mixed integer linear programs. It is based on the API of the ompr package and is also inspired by the architecture of Julia JuMP.

The goal is to provide a modelling package that can both be used in packages and also in interactive analyses. It also has a different architecture as the modelling layer modifies a central solver. That solver could be an interface to ROI or a shared pointer to a specific solver. Thus giving the option to directly communicate with the solver while still using an algebraic modelling framework.

This is currently work in progress and experimental - but working. I might merge it with ompr but it could also become the successor of ompr … not sure yet.

If you want to see the package in action take a look at the articles in the docs.

Happy to receive feedback!

Still under development. Anything can change

You can install the released version of RMPK from CRAN with:

remotes::install_github("r-opt/rmpk")

Linear Programming (LP)
Mixed Integer Linear Programming (MILP)
Mixed Integer Quadratic Programming (MIQP)
Mixed Integer Quadratically Constrained Programming (MIQCP)

✅ Algebraic modelling of mixed integer programming problems
✅ Integer, binary and continious variables
✅ Linear and quadratic constraints/objective
✅ Bindings to most popular solvers through ROI
✅ Support for character variable indexes
✅ Access row/column duals of Linear Programs
✅ Row generation through solver callbacks (e.g. for models with exponential many constraints)
🚧 Variable and constraint names
🚧 Initial feasible solutions
🚧 Almost as fast as matrix code
…

library(rmpk)
library(ROI.plugin.glpk)
set.seed(42)
solver <- ROI_optimizer("glpk")
v <- rnorm(10)
w <- rnorm(10)
model <- optimization_model(solver)
x <- model$add_variable("x", type = "binary", i = 1:10)
model$set_objective(sum_expr(v[i] * x[i], i = 1:10), sense = "max")
model$add_constraint(sum_expr(w[i] * x[i], i = 1:10) <= 10)
model$optimize()
model$get_variable_value(x[i])
#>    name  i value
#> 1     x  1     1
#> 2     x  7     1
#> 3     x  5     1
#> 4     x  8     0
#> 5     x  2     0
#> 6     x 10     0
#> 7     x  9     1
#> 8     x  3     1
#> 9     x  6     0
#> 10    x  4     1

rmpk supports all solvers that implement the MOI interface. It also comes with a ROI_optimzer that internally uses ROI and thus gives access to most popular solvers out of the box. Note that ROI has its own plugin system and you need to install these solvers separately in addition to ROIoptimizer.