Re: wip: functions median and percentile

2010/8/19 David Fetter <david@fetter.org>:
> On Thu, Aug 19, 2010 at 12:49:45PM -0400, Robert Haas wrote:
>> On Thu, Aug 19, 2010 at 11:33 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> > Greg Stark <gsstark@mit.edu> writes:
>> >> On Thu, Aug 19, 2010 at 11:59 AM, Pavel Stehule <pavel.stehule@gmail.com> wrote:
>> >>> I am sending a prototype implementation of functions median and
>> >>> percentile. This implementation is very simple and I moved it to
>> >>> contrib for this moment - it is more easy maintainable. Later I'll
>> >>> move it to core.
>> >
>> >> So if the entire result set fits in memory it would be nice to use the
>> >> O(n) Quickselect algorithm -- which would only be a small change to
>> >> the existing Quicksort code -- instead of sorting the entire set.
>> >
>> > That seems like rather a lot of added infrastructure for functions whose
>> > popularity is at best unknown.  I think we should KISS for the first
>> > implementation.
>>
>> +1.  I think the functions are useful, but the perfect is the enemy of the good.
>
> Percentile is already there as NTILE, a windowing function.  Median

I don't think it. The purpose is same, but usage is different -
aggregate functions can be used as window func, but window functions
cannot be used as aggregates.


> may be useful, but we pretty much can't just call it "median."
> Instead, we need to call it something like "left_median" or
> "arithmetic_median."

I put some time to deep searching in this area - and I talked with my
kolegues mathematican and everywhere use a just median. Some longer or
different name isn't barrier for me, but people can be confused.

Regards
Pavel

>
> Cheers,
> David.
> --
> David Fetter <david@fetter.org> http://fetter.org/
> Phone: +1 415 235 3778  AIM: dfetter666  Yahoo!: dfetter
> Skype: davidfetter      XMPP: david.fetter@gmail.com
> iCal: webcal://www.tripit.com/feed/ical/people/david74/tripit.ics
>
> Remember to vote!
> Consider donating to Postgres: http://www.postgresql.org/about/donate
>