Le 05/08/2010 22:32, Steven Watanabe a écrit :

AMDG

Ronan Trépos wrote:

...

For validity concerns, I'm trying to reproduce pseudo random numbers generated by the sofware R with boost::random library. The R default number generator is the Mersenne-Twister with period 2^19937 -1 .

1. In R I tried :

...

set.seed(13) runif(1)

0.7103224

2. With boost I tried :

boost::mt19937 rng(13); boost::uniform_01<boost::mt19937> zeroone(rng); std::cout << zeroone() << std::endl;

0.777702

Has someone already experienced that ? Is it possible to reproduce the same numbers ? Maybe the mismatch comes from the fact that generated integers are translated into [0;1)..

I don't think so. I think the difference is that R uses its own seeding algorithm.

Ok, thank you. I think I will give up. In R, concerning the MT generator, it is said : "The ‘seed’ is a 624-dimensional set of 32-bit integers plus a current position in that set." I understand that the 'seed' is actually the MT state. Calling set.seed in R creates the .Random.seed object which is a 626-dimensional set. (however I did not find what is the seeding algorithm used in set.seed). I think that the 624 last integers are the MT state and that the 2nd integer should be the current position in that set. In boost, I tried to construct the MT directly with the mersenne-twister(It& first, It last) constructor where It is an iterator on an integer vector containing .Random.seed values. I presume this constructor can be used to directly inititialise the MT state. Am I right ? However, this process did not work neither. Best regards, Ronan Trépos

In Christ, Steven Watanabe

_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

-- Trépos Ronan INRA unité BIA, équipe MAD, Toulouse http://carlit.toulouse.inra.fr/wikiz/index.php/Ronan_TREPOS tel : 05 61 28 53 34