Version française
Home     About     Download     Resources     Contact us    
Browse thread
Random questions
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: AUGER <Cedric.Auger@l...>
Subject: Re: [Caml-list] Random questions
Le Thu, 03 Dec 2009 15:48:37 +0100, Daniel Bünzli  
<daniel.buenzli@erratique.ch> a écrit:

> Hello Cedric,
>
> Thanks for your comments. Comments on your comments.
>
>> let rint () = (Random.bits () lsl 1) lxor (Random.bits ());;
>
> That was actually my first version. However I dropped it because I
> thought that generating a new random number by the interaction of the
> bits of two successive PRN could somehow change the underlying quality
> of the generator. Any thought ?

If it is the case, then the generation is of bad quality...
If you can predict the generated values of such combinations,
  then it is easy to predict the values of the base generator;
  which is then a bad generator.
(What I say is not valid for non reversible combination, such as "or" or  
"and",
  even if it implies that generator is of poor quality too;
  and not valid for complex operation, such as emulating the generator  
itself).
I am not expert in this domain, so it is better to directly ask on  
specialists.

>
>> if Random.bool is more efficient than Random.bits, the latter seems the
>> best; to be benchmarked
>
> Random.bool is Random.bits () land 0 = 1.
ok, so this function is not as interessant as I thought.

>
>>> 2) Generate an arbitrary int in [0;max] (bounds included)
>>>
>>> let random_uint ?(max = max_int) =
>>>  if max < 0 then invalid_arg "negative max" else
>>>  if max = max_int then Random.bits else
>>>  let bound = max + 1 in
>>>  fun () -> Random.int bound
>>
>> Seems correct.
>> I don't know if use of exception is more efficient or not, it is to try:
>>
>> let random_uint ?(max = max_int) () =
>>  try if max = max_int then Random.bits () else Random.int (max + 1)
>>  with Invalid_argument "Random.int" -> invalid_arg "negative max"
>
> You lose the ability to partially apply random_uint with the max and
> avoid the if on each call.
>
>
>> I think it is better to use the bool trick;
>> simpler and doesn't use Int32:
>>
>> let random_int ?(max = max_int) =
>>   if max < 0 then invalid_arg "negative max" else
>>   if max = 0 then 0 else
>>   if Random.bool ()
>>   then if max = max_int then Random.bits ()
>>        else Random.int max
>>   else negate_minus_1 (Random.int (max - 1)) (* inline? *)
>>
>
> Yes, thanks. Your version can also be reordered to be partially  
> applicable.
>
> It should also be noted that all our functions for [int]s don't work
> with the intended semantics on 64 bits platforms.

I noticed it, but I may have forgot to write it down.

>
>>> 5) Generate an arbitrary float in [0;max] (bounds included)
>>>
>>> let after_one = 1. +. epsilon_float
>>> let random_ufloat ?(max = max_float) =
>>>  if max < 0. then invalid_arg "negative max" else
>>>  fun () -> (Random.float after_one) *. max
>>
>> This function is less interesting than Random.float:
>> only less or as many as values as floats in [0.;1.] can be generated
>> so you will miss many values only to be able to produce max
>> (for example, you cannot 1+eps if you try random_ufloat ~max:2. ())
>
> The point you make maybe valid but unfortunately that's also the case
> for Random.float as this is the way it is implemented (generate a
> number in [0;1[ and scale it).
Ok, in fact it is reasonnable to do it since generating a random bit  
sequence (uniformly)
won't lead to a uniform law on floats, taking the exposant in  
consideration, will result in
more fine grained generator, but at a rather high cost.
and between 0 and 1 you can use int generation.

Furthermore, use of generated random floats is often done in [0;1] (as it  
is a probability).

>
> Best,
>
> Daniel
>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs


-- 
Cédric AUGER

Univ Paris-Sud, Laboratoire LRI, UMR 8623, F-91405, Orsay