Archives of the TeradataForum
Message Posted: Wed, 06 Jun 2007 @ 09:27:00 GMT
| Subj: || || Re: Minimal Stats Sampling and a well-informed |
| From: || || Geoffrey Rommel |
| ||The number of distinct values will end up being extrapolated upwards, depending on the percentage of rows sampled. If 2% of the rows were
sampled, then you could expect to see about 50 times more distinct values in the produced estimate....|| |
Carrie, this seems odd to me. Two percent is a large sample -- far more than is needed for a 99% confidence level. Why doesn't the sampler
simply accept what it finds? Is it because of a fear that the sample isn't "truly random"?