ROI.plugin.coinclp

Registers COIN-OR Clp with the R Optimization Infrastructure, through the coinclp package.

This revives ROI.plugin.clp, archived from CRAN on 2022-01-08 — not for any fault of its own, but because the clpAPI package it depended on had been archived six weeks earlier. The solver is now called "coinclp" rather than "clp", since ROI takes the solver name from the package name.

Installation

Clp itself is a system library; install it first (nothing to do on Windows, where Rtools supplies it).

# install.packages("remotes")
remotes::install_github("SamLovick/coinclp")
remotes::install_github("SamLovick/ROI.plugin.coinclp")

Use

Loading the package is the whole setup:

library(ROI)
library(ROI.plugin.coinclp)

op <- OP(objective   = c(143, 60),
         constraints = L_constraint(rbind(c(120, 210), c(110, 30), c(1, 1)),
                                    rep("<=", 3), c(15000, 4000, 75)),
         maximum     = TRUE)

res <- ROI_solve(op, solver = "coinclp")

solution(res)            # 21.875 53.125
solution(res, "objval")  # 6315.625
solution(res, "dual")    # 0.0000 1.0375 28.8750
solution(res, "aux")     # $primal row activities, $dual reduced costs
solution(res, "msg")     # the full coinclp result, including iterations

Constraints held as a simple_triplet_matrix, which is how ROI stores them, go to Clp in sparse form — nothing is densified on the way.

Controls

Control	Meaning
`log_level`	Clp’s message level, 0 (silent, default) to 4
`max_iterations`	iteration limit
`max_seconds`	time limit in seconds
`algorithm`	`"auto"`, `"primal"`, `"dual"`, `"barrier"`, `"barrier_nocross"`
`presolve`	run Clp’s presolve, `TRUE` by default
`primal_tolerance`, `dual_tolerance`	simplex tolerances
`scaling`	0 off, 1 equilibrium, 2 geometric, 3 auto, 4 dynamic

ROI_solve(op, solver = "coinclp", algorithm = "barrier", max_seconds = 30)

The names the 2018 plugin used — amount, verbosity_level, iterations, seconds — are still accepted, so scripts written for it keep working after changing the solver name.

Differences from ROI.plugin.clp

Duals, reduced costs and row activities are returned, not just the primal solution and the objective value.
Problems with no constraints, a dense constraint matrix, or an objective given as a plain vector are handled; the 2018 plugin errored on all three.
Status codes cover Clp’s full set, including “not solved”.
The algorithm, presolve, tolerance and scaling controls are new.

Licence

Eclipse Public License, matching COIN-OR and the original plugin. The code is independent of ROI.plugin.clp by Benoit Thieurmel (datastorm-open), whose control names are kept for compatibility.