Jeff Hoffmann <jeff(at)propertykey(dot)com> writes:
> does anybody on the list know anything about fractals & wavelets?
Now there's an interesting idea: regard the stats as a lossy compression
of the probability density of the original dataset. Hmm ... this
doesn't do anything for the problem of computing the pdf cheaply to
begin with, but it might help with storing it compactly in pg_statistic.
regards, tom lane