 |
 |
Archives of the TeradataForum
Message Posted: Thu, 20 Feb 2003 @ 13:32:46 GMT
| Subj: | | Re: Random Sampling |
| |
| From: | | Geoffrey Rommel |
| | I have been asked to provide a "truly random" sample of records from a table that exists in our Teradata. | |
| | While I have read the manual on SAMPLE and RANDOM I am unsure that these functions will provide a statistically valid sample. | |
Someone once told me that the algorithm used by Teradata is lrand48. This is a well-known algorithm that operates with 48-bit
numbers and I believe is considered pretty good. As for the sampling, I strongly suspect that they use the Waterman algorithm (Knuth,
Algorithm 3.4.2R).
The outputs of lrand48 probably meet all the tests you would care to meet (spectral, chi-square, etc.), but if your application demands
numbers not produced by an algorithm of any kind, you must go to a truly random source, such as
www.random.org.
--wgr
| |
