Thanks for pointing this out! In addition to not wanting a minimum, it's also bad to fix the number of samples. Then you know that the more you get sampled the

Thanks for pointing this out! In addition to not wanting a minimum, it's also bad to fix the number of samples. Then you know that the more you get sampled the less likely additional samples are. Might not matter for some applications but, as you say, it can subtly affect your behavior. The right way to do the sampling is as a Poisson process, ie, where the gaps between samples are drawn from an exponential distribution. Details: http://tagti.me http://tagti.me