So, for today's epic failure, I pulled a subset of data from work and tried building a cube from samples using a few different strategies. Most of the cube cells were aggregates of over 10,000 rows, so I was expecting them to converge pretty quickly. Nope. I have no idea what to make of this data. Missing even a single row can throw it way off. That is, of course, what the actuaries have been trying to tell me and why they don't really like the sampling approach.
Well, if it was an easy problem, they wouldn't give me a PhD for solving it. I'm meeting with my adviser tomorrow. Maybe he's got some ideas on it.
No comments:
Post a Comment