17 Jun
2015
17 Jun
'15
5:54 a.m.
AMDG On 06/16/2015 03:43 PM, Jonathan Jones wrote:
I've isolated this to a change in the implementation of normal_distribution between these two versions (mt19937 is producing the same sequence of numbers with the same starting seed).
My question is, is there any guarantee that this distribution is going to produce the same answers across different versions of Boost, given the same starting seed (zero in this case)?
No. The engines are guaranteed to produce consistent outputs, but there are no such guarantees for distributions. Note that implementations of std::normal_distribution are likely to produce different results from Boost and from each other, as well. In Christ, Steven Watanabe