|
|
|
|
|
by fanf2
748 days ago
|
|
I wrote about this at https://dotat.at/@/2022-04-20-really-divisionless.html You don’t need to worry about this for mapping hashes to buckets, because a hash has a limited width, whereas unbiased random numbers are based on an unbounded stream of random bits. In this article they are working with 128 bit hashes so they already have plenty of bits for bias not to be a problem. A single straightforward multiplication is fine. |
|