On Nov 1, 2012, at 23:49, robin wrote: > >> Here I see no evidence of clustering. > > There won't be. > All the data is uniformly distributed. > > Real data is not uniform.
OK. I'm starting to understand. But can you be more specific? Narrow down the definitions of "real" and "not uniform" For example, if I I choose the prime divison 37, advocated by JG, and my real data are "not uniform" but their density happens to correlate with a periodicity of 37, I will get adverse clustering with that modulus. I suppose JG's assertion (and yours) is a consequence of some postulate such as "'Real data' are unlikely to exhibit a periodicity of a(ny) prime number, but more likely to exhibit a periodicity which is a composite number." Is this intuitive (can you motivate me?) or empirical (can you cite the observations?) I grant that the non-uniformity of "real data" is the reason that compression techniques work. The Kolmogorov complexity of uniform data is itself. -- gil
