As a benchmark I ran the 10 million rating GroupLens data set, on a high CPU extra large instance. It generated recommendations for all 71K users in about 10 hours, costing a bit more than $8. That is a kind of worst case; you can recommend for only a subset of users or trade some accuracy for speed.
On Apr 3, 2009 9:49 PM, "Steve Lianoglou" <[email protected]> wrote: Hi Sean, > Incidentally for anyone that was interested, I did get an Amazon EC2 > image ready that reads an... Just out of curiosity, can you give an idea of how large your dataset was, how many cpus you used in "the cloud," how long it took, and a rough estimate of the EC2 cost? Thanks, -steve
