Don't have much time to write today, so I'll just mention that CISS seems to be converging quite nicely on the sample data. Actually, just stratifying the results greatly accelerated convergence using more conventional sampling, but when we get to the full data set (which, at 50 billion rows, is five orders of magnitude larger than the sample), simply stratifying won't do any better than the index lookups we're currently using. Hopefully CISS will, though it's going to be a while before I get this hooked to the full set.
Meanwhile, I still have the non-trivial task of deriving real prior and posterior distributions for my result. Without a mathematically defensible confidence interval, this thing is useless to the actuaries.
Full discussion and pretty graphs of convergence will also follow in the next few days.
No comments:
Post a Comment