using-other-solvers.Rmd

---
title: Using Other Solvers
author: Anqi Fu and Balasubramanian Narasimhan
date: '2018-05-15'
slug: cvxr_using-other-solvers
bibliography: ../bibtex/cvxr_refs.bib
link-citations: true
categories: []
tags: []
---


```{r prereqs, message = FALSE, echo = FALSE}
library(CVXR)
library(magrittr)
library(kableExtra)
```

## Introduction

The default installation of `CVXR` comes with several (imported) open source solvers:

- [ECOS](https://github.com/embotech/ecos) and its mixed integer
  cousin `ECOS_BB`
- [OSQP](https://www.osqp.org)
- [SCS](https://github.com/cvxgrp/scs)
  
`CVXR` can also make use of commercial solvers and other open-source solvers:

- The commercial solver [`MOSEK`](https://www.mosek.com/)
- The commercial solver [`GUROBI`](https://www.gurobi.com/)
- The commercial solver [`CPLEX`](https://www.ibm.com/analytics/cplex-optimizer)
- The open source linear and mixed integer programming package
  [`GLPK`](https://www.gnu.org/software/glpk/) via the
  [`Rglpk`](https://cran.r-project.org/package=Rglpk) package.
- The CBC solver via the R package
  [`rcbc`](https://github.com/dirkschumacher/rcbc). This is not yet on
  CRAN but can be installed directly from github.

Since these are optional, you have to install these packages
yourself. 

__NOTE__: `MOSEK` ,`GUROBI`, and `CPLEX` require licenses to use them
but free academic licenses are available for all three.

## Solver capabilities

Table 1 below shows the capabilities of various solvers.  Once the
appropriate packages are installed, a call to
`CVXR::installed_solvers()` will indicate what solvers `CVXR` is aware
of.


```{r, echo = FALSE}
check <- "\U2713"
d  <- tibble::tibble(
                  Solver = c("CBC", "GLPK", "GLPK_MI", "OSQP", "CPLEX", "ECOS", "ECOS_BB", "GUROBI", "MOSEK", "SCS"),
                  `R package` = c("[rcbc](https://github.com/dirkschumacher/rcbc)",
                                  "[`Rglpk`](https://cran.r-project.org/package=Rglpk)",
                                  "[`Rglpk`](https://cran.r-project.org/package=Rglpk)",
                                  "[osqp](https://www.osqp.org)",
                                  "[Rcplex](https://cran.r-project.org/package=Rcplex)",
                                  "[ECOSolveR](https://cran.r-project.org/package=ECOSolveR)",
                                  "[ECOSolveR](https://cran.r-project.org/package=ECOSolveR)",
                                  "[gurobi](https://www.gurobi.com/documentation/9.0/refman/ins_the_r_package.html)",
                                  "[Rmosek](https://cran.r-project.org/package=Rglpk)",
                                  "[scs](https://cran.r-project.org/package=scs)"),
                  LP = c(check, check, check, check, check, check, " " , check, check, check),
                  QP = c(" " , " " , " " , check, check, check, " " , check, check, check),
                  SOCP = c(" " , " " , " " , " " , check, check, " " , check, check, check),
                  SDP = c(" " , " " , " " , " " , " " , " " , " " , " " , check, check),
                  EXP = c(" " , " " , " " , " " , " " , " " , " " , " " , check, check),
                  MIP = c(check, " " , check, " " , " " , " " , check, check, check, check)
              )
knitr::kable(d, format = "html", caption = "Solver Capabilities") %>%
    kable_styling("striped") %>%
    column_spec(1:8, background = "#ececec")
                  
```

## Commercial Solvers

Pre-1.0 versions of `CVXR` used vendor python solver packages via
[`reticulate`](https://cran.r-project.org/package=reticulate), _not R
packages_. Version 1.x directly uses native R packages that have now
become available, resulting in a cleaner, more efficient interface.

Brief pointers on installation follow.

### Installing `MOSEK`

[MOSEK](https://www.mosek.com) provides an academic version that is
free of charge: one can obtain the free academic license after
registering. Once the license for the product has been activates, the
[Rmosek documentation](https://docs.mosek.com/9.1/rmosek/index.html)
provides all the details for installation. A quick check to ensure
things are working is to run the example:

```{r, eval = FALSE}
library(Rmosek)
example(mosek)
```

Once everything is working, you can check that `CVXR` recognizes the
solver; `installed_solvers()` should list `MOSEK`. Otherwise, rinse
and repeat until success.

### Installing `GUROBI`

[GUROBI](https://www.gurobi.com) also provides an academic version
that is free of charge. After registering, install the _Gurobi
Optimizer_ software and activate your license as necessary. The
[Gurobi
documentation](https://www.gurobi.com/documentation/9.0/quickstart_mac/r_interface.html#section:R)
provides details.

Once activated, you can check that `CVXR::installed_solvers()` lists
`GUROBI`. Otherwise, rinse and repeat until success.

### Installing `Rcplex`

[CPLEX](https://www.ibm.com/analytics/cplex-optimizer) is available in
a community edition. After installation, the
[`Rcplex`](https://cran.r-project.org/package=Rcplex) package needs to
be told where to find the libraries for linking. A sample session on
macOS is shown below for reference, where CPLEX version 22.1 was installed in the
standard `Applications` folder.

```{bash, eval = FALSE}
cplex_location <- "/Applications/CPLEX_Studio_Community221"
configure_args <- c(Rcplex = sprintf("--with-cplex-include='%s/cplex/include' --with-cplex-lib='-L%s/cplex/lib/x86-64_osx/static_pic -lilocplex -lcplex'", cplex_location, cplex_location))
install.packages("Rcplex", configure.args = configure_args)
```

### Example session

```{r}
CVXR::installed_solvers()
```

## Session Info

```{r}
sessionInfo()
```

## Source

[R Markdown](https://github.com/bnaras/cvxr_docs/blob/master/content/cvxr_examples/using-other-solvers.Rmd)

## References