alope107
alope107
@iamanigeeit Thanks so much, this was incredibly helpful! There are a few small typos in your dictionary though, where very a very similar looking character like ":" is used instead...
Ah, I think I get what's going on. In the linked article they use an underlying prng that has a range of [0,255]. With that, 107 is a significant portion...
To check the theory of the higher limits being the problematic ones, I did a quick test of the biased generator with a limit of 901100981: ``` bn::core::init(); auto random...
Looks much better to my eye! If I find some time later I could do some actual statistical tests if you're interested. Also, do you know if there's a way...
Sure thing, I'm busy for the rest of today but should be able to find some time tomorrow to automate this analysis a bit better and take a more thorough...
Ran a bunch of chi-squared tests for different sizes of limits, only detected anything significant for very large numbers using the old (biased) methods. Looks like the fix works perfectly!...