Sample performance

[Yomguithereal]

Currently, the sample function runs in O(k) (k being sample size) time but in O(n) (n being the size of the population). This can be awful for algorithms sampling large lists frequently (k-means, NN-descent etc...).

I developed some more efficients methods (having their drawback, however) in the pandemonium library (reservoir sampling still lacking but will soon be added). Does some of the library's method interest simple-statistics?

[tmcw]

Absolutely! Simple-statistics doesn't have much flexibility in sampling: I started with the Fisher-Yates approach because I could be comfortable with it being suitably random. Would love to add other methods, especially ones that are similarly random-enough and ones that work with streaming data.