Octopus max generations

Replies to This Discussion

Permalink Reply by Robert Vier on January 22, 2013 at 4:37

This depends on the problem you are working on, how its so-called 'fitness landscape' is composed. anticipating this is a rather tricky task, and is even made more difficult by multiple objectives (compared to single-objective optimization). hence, contemporary MO-search algorithms assume no knowledge of the design space and try to be designed as exploitative and robust as possible.

http://www.tik.ee.ethz.ch/sop/publicationListFiles/zitz2007a.pdf

this should give you a good overview of where EMO currently stands. in this sense I also must admit that SPEA-2 is slightly outdated regarding even exploitation of the search space, since a good set of solutions for the user

- is close to the real pareto front (check)

- is evenly distributed along its approximation of the pareto-front (check)

- has extents in the search space as big as possible

in the last point, SPEA-2 shows an internal bias towards specific solutions in some cases... the problem is a robust numeric indicator for all three targets, which at the moment (2013) seems to be a form of the hypervolume indicator. from march on, a research project i am doing will deal with this, and probably octopus is going to have some new algorithms built in additionally.

for now its just a matter how smart you set your objectives / the problem itself a.s.o..

PS do you know you can look, zoom and pan in the viewport of octopus?

holding the right mouse button while moving around rotates the scene, holding shift/ctrl while moving around zooms the thing. then you should be able to select everything you want.

Permalink Reply by jorge sainz de aja on January 22, 2013 at 5:02

thanks a lot

Permalink Reply by Fred Becquelin on January 22, 2013 at 5:37

Very nive papdr thank you!