Dne 12.4.2011 09:33, Claudio Freire napsal(a):
> I actually implemented a statistical system for measuring these kinds
> of correlations.
>
> It's complex, but it might be adaptable to pgsql. Furthermore, one of
> the latest projects of mine was to replace the purely statistical
> approach with SVCs.
You mean Support Vector Classifiers? Interesting idea, although I don't
see how to apply that to query planning, especially with non-numeric
inputs. Could you share more details on that statistical system and how
do you think it could be applied in the pgsql world?
> Too bad I won't be able to devote any time to that project before september.
I've been working on cross column stats for some time, and although I
had to put it aside for some time I'm going to devote more time to this
issue soon. So interesting ideas/comments are very welcome.
regards
Tomas