Skip site navigation (1)
Skip section navigation (2)
## Re: [HACKERS] Bad n_distinct estimation; hacks suggested?

### In response to

### Responses

### pgsql-performance by date

### pgsql-hackers by date

Quoting Andrew Dunstan <andrew(at)dunslane(dot)net>: > After some more experimentation, I'm wondering about some sort of > adaptive algorithm, a bit along the lines suggested by Marko Ristola, but limited to 2 rounds. > > The idea would be that we take a sample (either of fixed size, or > some small proportion of the table) , see how well it fits a larger sample > > (say a few times the size of the first sample), and then adjust the > formula accordingly to project from the larger sample the estimate for the full population. Math not worked out yet - I think we want to ensure that the result remains bounded by [d,N]. Perhaps I can save you some time (yes, I have a degree in Math). If I understand correctly, you're trying extrapolate from the correlation between a tiny sample and a larger sample. Introducing the tiny sample into any decision can only produce a less accurate result than just taking the larger sample on its own; GIGO. Whether they are consistent with one another has no relationship to whether the larger sample correlates with the whole population. You can think of the tiny sample like "anecdotal" evidence for wonderdrugs. -- "Dreams come true, not free." -- S.Sondheim, ITW

- Re: [HACKERS] Bad n_distinct estimation; hacks suggested? at 2005-04-25 20:43:10 from Andrew Dunstan

- Re: [HACKERS] Bad n_distinct estimation; hacks suggested? at 2005-04-27 13:43:31 from Andrew Dunstan

Next: From:Greg StarkDate:2005-04-27 05:59:30Subject: Re: [PERFORM] Bad n_distinct estimation; hacks suggested?Previous: From: Tom LaneDate: 2005-04-27 04:14:36Subject: Re: [PERFORM] Bad n_distinct estimation; hacks suggested?

Next: From:Greg StarkDate:2005-04-27 05:59:30Subject: Re: [PERFORM] Bad n_distinct estimation; hacks suggested?Previous: From: Christopher Kings-LynneDate: 2005-04-27 04:29:39Subject: Re: Disable large objects GUC