Developing an R package using Julia for a statistical model

Specific Domains Statistics
1

Hi all,

I would like to know what is the current state of using Julia as a backend for an R package. I writing a code for a statistical model, but I aware that most of the people will use R to run those analysis so I would like to know if is possible now days to have a R package with Julia running in the backend.

If is not possible, at the moment so it might not be worth the pay off of trying to implement in Julia from the scratch.

Thank you all,

2 Likes
2

There was some development of compiling Julia code to small binaries which R can then call. It was in another thread. I expect to see development over the next 2 years or so

3

yes, most of the things that are so are more related with small chunks of code being generated in R, but nothing like could be used in a R package

4

I don’t know much about R but there is a rather high-profile Python package for symbolic regression which is just a front end for Julia:

Maybe @MilesCranmer would know some R equivalents

4 Likes
5

There is at least one differential equation solver available in r that is entirely based on Julia

Julia will become like rcpp in a few years

1 Like
6

Do you have the link of this diff. equation solver package? I would be interested and would be very useful if I can explore the git repository

7

GitHub - SciML/diffeqr: Solving differential equations in R using DifferentialEquations.jl and the SciML Scientific Machine Learning ecosystem?

3 Likes
8

Yeah. I found it by googline “differential equation solver r package coded in julia”

1 Like
9

What’s their approach to Julia installation? Is it handled within pip or conda on the Python side?

10

I think everything is handled with this dual Julia/Python binding package:

1 Like
11

It’s already possible to call Julia from R (as other have suggested for package development), and done, and note also RCall.jl for calling in other direction, with its R mode.

Note also:

And upcoming juliac compiler will likely make this even more compelling, but it’s strictly not needed, current tools good enough.

I think everything is handled with this dual Julia/Python binding package

That PythonCall package is of course Python specific, but it relies on CondaPkg.jl which isn’t. I’m not sure, maybe, probably, it helps for R users installing Julia dependencies?

12

See this thread. With the work that has gone into small binary compilation of Julia code, it now seems possible to create a dynamic library that you can just call in R. You may want to reach out to the authors there to see if you can help with this.

2 Likes
13

As described in that paper, the R package fwildclusterboot uses the Julia package WildBootTests.jl as a back end. For some problems the Julia backend is optional (but faster). For others, it is required. The Stata package boottest can use the same back end.

For sure, once juliac is more official, I’ll try to move to that so that users need not install Julia.

14

I am expecting juliac built small binaries to work for this purpose by Julia 1.12, but no later than Julia 1.13. Probably, we’ll need some tooling to then make it all work well together.

-viral

10 Likes
15

Note:
https://cran.r-project.org/web/packages/JuliaCall/readme/README.html

Or you can do

library(JuliaCall)
julia_setup(installJulia = TRUE)

which will invoke install_julia automatically if Julia is not found and also do initialization of JuliaCall.

I don’t know if there are any problems, if it’s as smooth as calling Julia from Python (PythonCall.jl can also install Python automatically) using that other package with same name:

JuliaCall manages its Julia dependencies using JuliaPkg. It will automatically download a suitable version of Julia if required.

I’m not sure why…:
https://cran.r-project.org/web/packages/fwildclusterboot/index.html

Package ‘fwildclusterboot’ was removed from the CRAN repository.
Formerly available versions can be obtained from the archive.[that’s a link at target, that didn’t copy-paste.]
Archived on 2024-05-29 as issues were not corrected in time.

[Note, in Julia you can’t remove packages from the index, that would be a security irsk, as happened for JavaScript…]

16

Yeah. I was think something more akin to Rcpp in R

17

This is often a problem in corporate environments because install stuff off the net is not allowed. So need something more suited

18

Yes, as @ xiaodai says, while ideally Julia installation should be smooth as butter, dependence on it is still a source of fragility. More things can go wrong if Julia is required than if you can just deliver a shared library to the user’s file store. There may also be some speed benefit, at least the first time the library is called in a session.

fwildclusterboot uses the JuliaCall package you referenced, or some comparable package.

19

Hi, I am in the process to develop Rulia to make some R package “wrapping” julia package. It is an alternative of JuliaCall which depends on Rcpp. My goal is to replace C++ by Julia to boost R code.

1 Like
20

I think nobody highlighted that there are 2 very different approaches to the problem:

  1. On one side an R package can depend on JuliaCall and when installed, it can install Julia and the desired Julia package that it wraps in the background. This I believe is the approach taken by current R wrappers to Julia packages.
  2. On the other side, the Julia package can be “compiled” in an lib/executable and the R package interface this directly (using the C API ??). This is way more “frontier” approach. For example, the “compilation” of a Julia package in a small binary that doesn’t take everything with it has been only very recently added to Julia 1.12(that still needs to be released…).