Nah, scratch that too. The simple version of this idea doesn't scale, and I was unable to get the current version to run at all significantly differently in speed. It's just good as-is.
Now there is a non-distributed similarity implementation that matches what this does, which was the original question. Sean On Wed, Apr 28, 2010 at 7:14 PM, Sean Owen <sro...@gmail.com> wrote: > Actually scratch that patch I sent over. I see the trick now that > makes the existing approach quite good. I think I can make a version > that preserves that trick and still streamlines the processing. I will > benchmark and report back if successful. > > On Wed, Apr 28, 2010 at 3:20 PM, Sean Owen <sro...@gmail.com> wrote: >> Sorry, typo, that's what I meant. yes the difference isn't *that* large! >> It may be worse in practice since you have a few users with very many prefs. >> It may also be beneficial to simply have one fewer phase and throw >> around less data. I will also try to benchmark since really that's the >> only way to know. >> >
