Correlated data: block bootstrap

[carstenbauer]

Great package!

AFAICS, the package is only providing online algorithms for uncorrelated data (some features can be applied to correlated data as well, of course).

Is there any plan to support correlated data specifically as well? Is this even possible efficiently? (I have little experience with online statistics.)

Concretely, I'd love to see a block bootstrap or block jackknife based confidence interval estimation that considers correlations by pre-binning that data. I guess it's only possible in O(logN) memory though.

Maybe you can share your plans/thoughts on this?

[joshday]

I think the only OnlineStat for correlated data is AutoCov. Also, the weighting allows you to essentially do moving window calculations with any OnlineStat.

I don't have any plans for work in this area, but if people give me papers with algorithms to implement I would add them when I have time. I also would accept PRs!