Optimization packages for R

Question

Does anyone know of any optimization packages out there for R (similar to NUOPT for S+)?

downvote for 'this question does not show any research effort' — Spacedman, Jan 06 '13 at 09:51
I "could" have made this a long question talking about the details of what I needed and how I had investigated linprog and found it wanting. Or I could just ask a very simple question. I stand by that decision. BTW, while I disagree with your reasons for downvoting me, at least you had the decency to give a reason. Thank you for your courtesy. — wcm, Jan 09 '13 at 21:02
If you'd told us where you had looked then that would have shown some research effort - even a simple "i searched google or cran for ' optimisation' " would have helped. We shouldn't have to point people to cran task views... And if you'd found linprog wanting, why accept the answer that says "I've used linprog"? — Spacedman, Jan 09 '13 at 21:42
It appears 3rd in google search under "linear optimization r", meaning we should be practical about it and use it as a reference for future searchers. — marbel, Feb 11 '14 at 23:43

score 20 · Answer 1 · edited Jan 06 '13 at 09:52

20

R has many, many packages for optimization; check the CRAN Task view on Optimization: http://cran.r-project.org/web/views/Optimization.html. Of course, for nonlinear programs, there is optim(), which is standard and includes Broyden-Fletcher-Goldfarb-Shanno's algorithm, and Nelder-Mead. It's a good first start.

edited Jan 06 '13 at 09:52

Spacedman

92,590
12
140
224

answered Jul 27 '09 at 21:23

gappy

10,095
14
54
73

score 7 · Answer 2 · answered Jan 06 '13 at 04:55

Try lpSolve with R.

A simple example:

# Maximize 
#   x1 + 9 x2 +   x3 
# Subject to: 
#   x1 + 2 x2 + 3 x3 <= 9
# 3 x1 + 2 x2 + 2 x3 <= 15
f.obj <- c(1, 9, 3)
f.con <- matrix(c(1, 2, 3, 3, 2, 2), nrow = 2, byrow = TRUE)
f.dir <- c("<=", "<=")
f.rhs <- c(9, 15)

lp("max", f.obj, f.con, f.dir, f.rhs)
lp("max", f.obj, f.con, f.dir, f.rhs)$solution

score 7 · Answer 3 · answered Jan 06 '13 at 05:13

you should also try the Rglpk package solve LP problems with GLPK (GNU Linear Programming Kit).

An example:

## Simple linear program.
## maximize:   2 x_1 + 4 x_2 + 3 x_3
## subject to: 3 x_1 + 4 x_2 + 2 x_3 <= 60
##             2 x_1 +   x_2 +   x_3 <= 40
##               x_1 + 3 x_2 + 2 x_3 <= 80
##               x_1, x_2, x_3 are non-negative real numbers

obj <- c(2, 4, 3)
mat <- matrix(c(3, 2, 1, 4, 1, 3, 2, 2, 2), nrow = 3)
dir <- c("<=", "<=", "<=")
rhs <- c(60, 40, 80)
max <- TRUE

Rglpk_solve_LP(obj, mat, dir, rhs, max = max)

R output:
(Note that $status an integer with status information about the solution returned. If the control parameter canonicalize_status is set (the default) then it will return 0 for the optimal solution being found, and non-zero otherwise. If the control parameter is set to FALSE it will return the GLPK status codes).

$optimum
[1] 76.66667

$solution
[1]  0.000000  6.666667 16.666667

$status
[1] 0

score 6 · Answer 4 · answered Jan 12 '09 at 16:20

6

Linprog, mentioned by Galwegian, focuses on linear programming via the simplex algorithm. In addition you may be interested in fPortfolio if you are doing portfolio optimization.

answered Jan 12 '09 at 16:20

JD Long

59,675
58
202
294

score 4 · Answer 5 · answered Dec 11 '08 at 14:06

4

I have used linprog for linear problems in the past.

answered Dec 11 '08 at 14:06

Galwegian

41,475
16
112
158

+1 and thanks a lot. I haven' touched any optimization stuff since my TurboPascal days. There were a whole bunch of other optimization packages listed in the packages page (http://cran.r-project.org/web/packages/). – wcm Dec 11 '08 at 15:27

score 3 · Answer 6 · answered Nov 03 '21 at 05:51

Another package is ompr. An advantage of this package is there are many solvers that can be used and binary, continuous, integer all variables can easily be added. A simple example:

library(tidyverse)
library(ompr)
library(ompr.roi)


model <-  MIPModel() %>%
  add_variable(x1, type = "integer") %>%
  add_variable(x2, type = "integer") %>%
  set_bounds(x1, lb = 0) %>%
  set_bounds(x2, lb = 0) %>%
  set_objective(x1 - x2, "max") %>%
  add_constraint(x1 + 2*x2 <= 150) %>%
  add_constraint(x1 >= 30) %>%
  add_constraint(x2 >= 40)

Solving with glpk:

library(ROI.plugin.glpk)
result <- solve_model(model, with_ROI(solver = "glpk", verbose = TRUE))

get_solution(result, x1)
get_solution(result, x2)

It can also be solved with other solvers like symphony where the gap_limit can be set in case the problem is complex and will take many iterations to converge:

library(ROI.plugin.symphony)
result <- solve_model(model, with_ROI(solver = "symphony",
                                      verbosity=-1, gap_limit=1.5))

score 2 · Answer 7 · answered Jan 11 '17 at 01:47

2

I like Gurobi. It's very expensive for a license, but it can be obtained through many universities. See here http://www.gurobi.com/products/modeling-languages/r

answered Jan 11 '17 at 01:47

beddotcom

447
3
11

score 2 · Answer 8 · answered Jun 20 '19 at 13:44

2

Have a look on the NLoptr package. It has quite an extensive documentation with examples and lots of algorithms to choose from, depending what problem you are trying to resolve (etc. linear, non-linear, constraint)

answered Jun 20 '19 at 13:44

nba2020

618
1
8
22

score 0 · Answer 9 · answered Jul 02 '23 at 06:00

0

Ideally inject a dynamic momentum in nlopt or any optimizer, involving inertia and viscosity. Does not seem available in .R

answered Jul 02 '23 at 06:00

jean valayer

1

Optimization packages for R

9 Answers9