Sean,
Any progress?
I have, in the meantime, written a hardcoded 1+1 ES that does strict
hillclimbing. It works as expected (although with unexpected results).
Cheers,
Steve
On Sep 18, 2005, at 3:49 PM, Sean Luke wrote:
> Hi Steve. If your information is correct, it looks to me that the
> problem is with QuickSort not being a stable sort. ECJ's ES library
> first sorts the population based on fitness, then uses the first
> lambda
> individuals in the population array as the new parent pool.
> Because it
> uses QuickSort, equal individuals do not necessarily sort the same way
> in the pool.
>
> ECJ's minimum version of Java is now 1.2, meaning we can probably toss
> out our QuickSort class and use java.util.List's sort routines,
> including stable sorts based on MergeSort. It'll take us a few days,
> but let us see if we can get an upgraded version of at least the ES
> package for you fairly soon.
>
> Sean
|