Re: [SQL] More fun with random selects

Darren Greer <dgreer(at)websightsolutions(dot)com> writes:
> For example, lets say I want a random 125 rows from an existing table. Is
> there an easy way I can do that where I actually am getting a decent sample
> from the table?

I don't think there's any way to do that with a simple SQL command.

It's possible to draw exactly N items from a set fully randomly if you
are willing to write a little code. The algorithm looks like this:

itemsRemaining = size of set (# rows in table);
itemsStillNeeded = N (# rows wanted, 125 in your example);
foreach (item in set)
{
generate random value X that is 'true' with probability
itemsStillNeeded / itemsRemaining;
if (X)
{
emit current item as a selected item;
decrement itemsStillNeeded;
if (itemsStillNeeded == 0) done;
}
decrement itemsRemaining;
}

The random choice is typically done with a random number generator
that generates output G between say 0 and M; you make X 'true' if
G <= M * itemsStillNeeded / itemsRemaining.

regards, tom lane