Archives of the TeradataForum
Message Posted: Thu, 16 Apr 2009 @ 14:53:04 GMT
| Subj: | | Re: Collecting Stats - Column or Index |
| |
| From: | | Victor Sokovin |
| | I'd accept this a lot quicker if it wasn't for the fact that statement immediately above it suggests Collecting Stats on (ColA,COlB,ColC)
is exactly the same as Collecting Stats against those columns individually, which I'm pretty sure is nonsense !! | |
Of course it is! All of us who have struggled through the multivariate statistics lectures must remember how complexity level rises compared to
the single variable case.
Individual column stats above - it's like a projection on one axis. You have three projections. They tell you something about the three-
dimenstional system but not very much at all.
More terminology and examples in this article:
en.wikipedia.org/wiki/Multivariate_normal_distribution
Now, this is as far as the underlying mathematics is concerned. The other thing is what exactly was implemented in Teradata under the name of
multi-column statistics. My latest understanding is that the implementation was not a success and it did not match the original user manuals
describing it (there should be a few threads in the archives of this Forum confirming this). Until a better implementation is in place we should
probably forget about the current one. Not worth our time.
Victor
| 
