Re: RNGs in Guile

On 5/28/26 12:25 AM, Tim Meehan wrote:
> I think that (ice-9 random) has (random:normal). That would generate a 
> Gaussian random variate with mean zero and standard deviation 1.
>
> On Wed, May 27, 2026 at 7:12 AM Zelphir Kaltstahl <[email protected]> 
> wrote:
>
>     On 5/27/26 7:19 AM, λx.x wrote:
>     > this is somewhat tangential; i am not necessarily talking of hashing,
>     but more
>     > cryptographic functions shipped with Guile as a whole.
>     >
>     > if Guile is going to ship with a cryptographic hash function, what about
>     other
>     > cryptographic functions?  Are we satisfied with the RNG interfaces
>     provided in
>     > core and SRFI 27?  whether they RNGs provided are cryptographically secure
>     > does not seem to be documented in the manual, at least not explicitly.
>
>     What I found lacking are functions normal/Gaussian distributed random number
>     generation. Last time I checked there was no such thing. Generating a normal
>     distribution from uniform distributions can be mathematically quite
>     challenging.
>     So many algorithms for approximation, and many of them requiring deep
>     mathematical understanding to understand when one such algorithm would yield
>     good enough results and when not, and what parameters to tweak to make it
>     suitable and so on and on. Or blindly copying without understanding and just
>     praying. Not very keen on having to improvise something like that, with
>     limited
>     mathematical understanding and Wikipedia being an impenetrable wall of
>     math for
>     such algorithms.
>
>     I went as far as checking NumPy, how normal distributed random numbers are
>     generated there, but the code sucks so hard, it is also impenetrable for
>     someone
>     not knowing the mathematical formula it tried to express and understanding
>     that
>     in turn. One letter variables or abbreviations everywhere, that no one other
>     than mathematicians will be able to interpret, with no regard for
>     readability at
>     all. Just like what one would expect a mathematician without any software
>     development experience to write : )
>
>     Having a good (and readable! perhaps with references!) algorithm for that
>     in the
>     standard library, or in a supported SRFI, or as an extension to a supported
>     SRFI, would be great. And the docs should of course state what purposes it is
>     useful for and whether it is cryptographically secure or not.
>
>     Not that Python necessarily is the yardstick for all the things, but here
>     is its
>     standard library: https://docs.python.org/3/library/random.html#random.gauss.
>
>     Best regards,
>     Zelphir
>
>     -- 
>     repositories: https://codeberg.org/ZelphirKaltstahl
>
Damn, how did I not find that, when I had the need. Feeling a bit stupid now. 
Thanks for telling me.

Just out of curiosity, how did you learn about that function existing? Just 
through procedure index of the manual and searching for "random" or "normal"?

Best regards,
Zelphir

-- 
repositories:https://codeberg.org/ZelphirKaltstahl