I worked out the exact block variance formula for the general case, taking into account the differences in the partition sizes. It's pretty nutty, but actually reduces quite nicely due to the fact that we know the partition sizes up front. Therefore, much of the formulas can be computed at load time and just looked up in a table at query time. I'm calling this a big win on two fronts: 1) Lots of math 2) actually computes quickly.
Yay for that.
No comments:
Post a Comment