Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

Interesting that the percentage difference between the winning solution with 107 blended approaches and the close-by solutions using only one approach was on the order of a fraction of a percent in improvement.


view as:

Yet another real-life application of the 80/20 rule....

In computational biology (my world), it's routine for an initial study to account for most of the eventual success of a method, with many man-years of work dedicated to improving success rates only by tiny fractions.

(As an aside, if you're an ambitious researcher, the implication of this is clear: don't waste your time refining things. Rapid testing of ideas is more important than squeezing out that last 1% of performance improvement. Iterative refinement is research death.)


Do you have a reference describing the single approach?

If you click around the links on the leaderboard [http://www.netflixprize.com/leaderboard], a lot of folks describe the basics of their approaches.

There are also a few entries (like "When Gravity and Dinosaurs Unite") that are explicit blends between two single approaches from different teams.


Legal | privacy