Re: Cross-column statistics revisited

Martijn van Oosterhout <kleptog(at)svana(dot)org> writes:
> Just a note: using a multidimensional histograms will work well for the
> cases like (startdate,enddate) where the histogram will show a
> clustering of values along the diagonal. But it will fail for the case
> (zipcode,state) where one implies the other. Histogram-wise you're not
> going to see any correlation at all

Huh? Sure you are. What the histogram will show is that there is only
one state value per zipcode, and only a limited subset of zipcodes per
state. The nonempty cells won't cluster along the "diagonal" but we
don't particularly care about that.

What we really want from this is to not think that
WHERE zip = '80210' AND state = 'CA'
is significantly more selective than just
WHERE zip = '80210'
A histogram is certainly capable of telling us that. Whether it's the
most compact representation is another question of course --- in an
example like this, only about 1/50th of the cells would contain nonzero
counts ...

regards, tom lane