0

I am using MKL Intel for random number computation. I use Quasi random algo. Also, it seems to me that number that are generated are not random at all. I understand that Quasi random algo will provide sampled points that maximize space occupation. Thus, they don't need to be perfectly unexpected and random, but only to cover the space properly.

I would like to use this number in a random fashion, not to have them ordered on an expected way, so that before using these 'random' points, I permute them.

Am I not using MKL library properly? Or does it sound normal, because, as I mentionned, points are to cover whole space, not to be perfeclty random?

If some MKL or quasi random gen. users there, advice welcome.

Regards.

kiriloff
  • 25,609
  • 37
  • 148
  • 229

1 Answers1

1

In theory random permutation of any uniformly distributed sequence should result in random sample. Take 1, 2, 3, ..., n, for example, and randomly permute it, and you get random uniform sample.

The quality of random permutation algorithm becomes vital to the quality of the resulting sample.

At the same time I wonder why you need quasi-random RNG from MKL at all. To me it looks a bit an overkill. If you need good pseudo-random sequence then it might make sense to use other RNGs from MKL, e.g. MT19937 or MT2203 or SFMT19937, which are fast and are of high quality. Can you please explain why you need a quasi-random sequence as the base for permutation?

Regards, Sergey

Sergey Maidanov
  • 26
  • 1
  • I dont need random generator as a base for permutation. This is not the point here. I am using MKL quasi random generator, and I notice that it returns always numbers in the same sequence (when I sample same given interval for instance). I wonder whether this behavior is expected for such a generator (which is not 'random' in proper sense), or whether I am not usnig MKL properly here. thanks – kiriloff May 15 '12 at 10:42