On Mon, Mar 21, 2016 at 2:39 PM, Alexander Korotkov <
> On Fri, Mar 18, 2016 at 1:20 PM, Dean Rasheed <dean.a.rash...@gmail.com>
>> Probably a better URL to give is
>> http://www.adellera.it/investigations/distinct_balls/ which has a link
>> to the PDF version of the paper and also some supporting material.
>> However, while that paper is in general very clear, I don't think it
>> gives a very clear explanation of that particular formula, and it
>> doesn't explain what it represents. It merely says that if "i" can be
>> ignored "for some reason (e.g. i << Nr)", then that formula is an
>> approximation to the exact "without replacement" formula, which is the
>> subject of that paper.
>> But actually, that formula is the exact formula for the expected
>> number of distinct values when selecting *with replacement* from a
>> collection where each of the distinct values occurs an equal number of
>> times. So I think we should say that.
>> Perhaps something along the lines of:
>> * Update the estimate based on the restriction selectivity.
>> * This uses the formula for the expected number of distinct
>> * when selecting with replacement from a collection where
>> each of
>> * the distinct values occurs an equal number of times (a
>> * distribution of values). This is a very close
>> approximation to
>> * the more correct method of selecting without replacement
>> * the number of distinct values is quite large --- for
>> * see http://www.adellera.it/investigations/distinct_balls/.
>> * Additionally, it also turns out to work very well even
>> when the
>> * number of distinct values is small.
> Thank you for work on this patch. The formula you propose and explanation
> look great!
I think you should send a revision of patch including comments proposed by
I'm switching patch status to waiting on author in commitfest.
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company