Online variance and SD

[superbobry]

Algorithm is described here.

[superbobry]

An even better idea would be to abstract ecirca_type in to a class-like structure and allow creating stacks of various summary statistics (since none of the "types" are mutually exclusive). For instance we might have:

ecirca:new(256, [mean, variance, max])

What do you think?