问题:

I am interested in exploring how R can handle data out-of-memory. I've found the bigmemory package and friends (bigtabulate and biganalytics), but was hoping that someone could point me to a worked out example that uses file backing with these packages. Any other out-of-memory tips would also be appreciated.

回答1:

Charlie, just email Mike and Jay, they have a number of examples working around the ASA 'flights' database example from a year or two ago.

Edit: In fact, the Documentation tab has what I had in mind; the scripts are also on the site.

回答2:

Take a look at "CRAN Task View: High-Performance and Parallel Computing with R". There is a chapter "Large memory and out-of-memory data" where severel solutions are mentioned. For example package ff.

回答3:

Any other out-of-memory tips would also be appreciated.

I frequently work with large datasets. Even though my code has been optimized, I still launch Amazon EC2 instances from time to time because it gives me access to far more resources than on my desk. For example, an instance with 26 ECUs, 8 cores, and 68 gigs of RAM only costs about a $0.80-1.00 per hour (spot instance pricing).

If that seems reasonable, you can launch a public machine image that already has R and do this job in no time.