On 23-03-2013 10:15, Andres Freund wrote:
> I just want to mention that unless you skew the statistics for the individual
> tables from their empty/default state this mostly measures a pretty degenerate
> case where optima are very rare and not very differentiated. Thats a useful
> thing to test, but not to have as the target to optimize for.
> So it might be interesting to run that thing with some table
> stats/contents stats set up.
Yes, the search space obtained from this experiment may be very simpler
than a real case. Beyond this experiment, I can construct classical
queries used to evaluate this kind of algorithm, as stars, cliques,
chains and cycles. Beyond these queries I have no idea how can I further
test it.
Regards,
Adriano Lange