Package with "big" data - how should I make data set available and conform to CRAN's requirements?

tiagocabaco · June 22, 2020, 9:02am

Hi everyone,

I am developing a package with a data set I scraped from a website and I would like to make it available on CRAN.

However, the compressed data is already higher then 30 mb.

I was wondering, what would be the recommended way to include the data in my package?

My idea was host to the data on Github, for example, and then have a function in the package to download the data file. However, I have no idea if this is a recommend practice in these types of situations.

Thank you for the help!

jlacko · June 22, 2020, 9:15am

When facing a similar issue I resolved it in the way you describe (i.e. a hosting the dataset online - Amazon S3 in my case - and having a function serve it).

In my case this was suggested by CRAN maintainers themselves (in a comment along the lines "take your huge data elsewhere" )

Some things for your consideration:

it is likely the dataset will be downloaded a lot, even for niche packages (a lot of CRAN traffic is CI testing), this may impact your choice of hosting provider
CRAN does not really care where your external data is stored, but expects you to access it via https
CRAN discourages you from caching the dataset locally (tempdir is OK, but that is the limit)
bear in mind that utils::download.file() can have platform related issues, resulting in your file being unreadable; I have found curl::curl_download() more reliable (your mileage may vary)

tiagocabaco · March 21, 2021, 1:20am

Thank you very much for your response.

Also, thanks for the very useful advice. I had no idea about utils::download.file() could be problematic, for example.

jlacko · December 3, 2020, 5:45am

Glad to be of service!

I have only learned about the download.file issues myself the hard way = by having a download fail for a package already released (I am on Linux and the specifics of Windows are sometimes lost on me).

It can be tweaked by specifying the method and mode arguments of the function call (instead of relying on auto mode), but having got burned I opted for cURL instead.

system · June 29, 2020, 10:29am

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.

EconomiCurtis · July 14, 2020, 12:14am

maelle · July 14, 2020, 7:11am

Two other ideas from a recent blog post I wrote on the R-hub blog.

you can cache data locally in an app dir. as long as the user is asked for permission; as to how make the data available and download it, as mentioned earlier in this thread there are many options. Useful packages: rappdirs, hoardr. New R function: tools::R_user_dir() (from R4.0, for older R one would need the backports package).
you could distribute the data in a non CRAN package this way it is downloaded by installing the package and cached in the user library.

tiagocabaco · August 4, 2020, 11:49am

Thank you for your response! Very useful! I ended up setting a remote installation function that the user calls when the package is installed.