Re: [Chicken-users] srfi-27 - producing pseudo and truly random numbers over a uniform distribution

[Kon Lovett]

On Oct 28, 2007, at 2:55 PM, Terrence Brannon wrote:

> In addition to the bug I filed on this egg - http://trac.callcc.org/ 
> ticket/346
>
> I have a question about common usage of this module.
>
> I would like to produce one function, roll-fixed, which produces
> pseudorandom numbers in the documented fashion of random-real and
> random in SRFI-27. Such that the sequence of randoms (over a uniform
> distribution) is predictable on each new startup of the intepreter:
>
> (define *fixed-seed* (expt 7 5))
>
> (define roll-fixed
>   (let ([side-effect (set-seed *fixed-seed*)])
>      (lambda (n)
>         (if (zero? n)
>             (random-real-based-on-fixed-seed)
>             (random-integer-based-on-fixed-seed n)))))

(use entropy-fixed)

; A fixed seed entropy struct

(define fixed-seed
  (make-procedure-entropy-structure
   (let ([*fixed-seed* (expt 7 5)]) (lambda () *fixed-seed*))
   "fixed seed: (expt 7 5)" ) )

; This uses the 'current-random-source-structure'

(define (roll-fixed n)
  (if (zero? n)
      (random-integer)
      (random-real) ) )

; As default entropy:
(current-entropy-source-structure fixed-seed)
(random-source-randomize! default-random-source)

; Or:
(random-source-randomize!/entropy default-random-source fixed-seed)

Also, the random-state of a random-structure can be saved & restored.  
So each run can load a file w/ a random-state before getting a random  
sequence.

Note, until the new release is posted 'entropy-fixed' & 'random- 
source-randomize!/entropy' are unavailable. 'entropy-fixed' is in the  
posted egg but is not in the .setup (I forgot it).

> Additionally, I would like a function, roll-truly-random, which again
> can produce reals or integer randoms, but whose sequence of randoms
> (over a uniform distribution) is different across each new startup of
> the interpreter.
>
> (define roll-truly-random
>   (let ([side-effect (set-seed (truly-random-seed))])
>      (lambda (n)
>         (if (zero? n)
>             (random-real-based-on-truly-random-seed)
>             (random-integer-based-on-truly-random-seed n)))))

What is '(truly-random-seed)'?

Currently the entropy structures 'dev-random-entropy', 'dev-urandom- 
entropy' are defined for unix & 'crypt-random-entropy' for windows,  
and 'current-seconds-entropy' (for both platforms).

Note that SRFI-27 does not define entropy sources - it just assumes  
there is one.

> I'm sorry to beg for such hand-holding, but it seems like the Issues
> section of the docs does admit that this library could use a few good
> examples. It also mentions that the API creates something of a hurdle
> for my desire to have two distinct random structures available in two
> distinct functions.

You cannot have distinct random structures but you can have distinct  
random sources. (Well, you can, MWC is distinct from MOA but I doubt  
you want to define your own random number algorithm.)

(use mwc mrg32k3a moa)

(define my-mwc-source ((MWC 'make-random-source)))
(define my-moa-source ((MOA 'make-random-source)))
(define my-mrg-source ((MRG32k3a 'make-random-source)))

(define my-mwc-random-integer (random-source-make-integers my-mwc- 
source))
(define my-mwc-random-real (random-source-make-reals my-mwc-source))
...
(random-source-randomize!/entropy my-mwc-source fixed-seed)
(my-mwc-random-integer 23) ; [0 23)
...

(use srfi-27-distributions)

(define my-mwc-random-integers (make-uniform-random-integers 22 15 1  
my-mwc-source))
(my-mwc-random-integers) ; [15 22] by 1

SRFI-27 was written assuming only one pseudo-random algorithm & a  
simple module system. While I could have extracted the random source  
agnostic procedures it seemed better to just follow the spec. In any  
case a random structure is open upon load so these procedures are  
always available.

> _______________________________________________
> Chicken-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/chicken-users

Best Wishes,
Kon