objective function

[gusl]

I'm looking for some easy-to-program, multimodal objectives over the {0,1}¹⁰ hypercube in which to test an optimization algorithm.

Any suggestions?

Comments

[the-locster.livejournal.com]

Well if you want to go 11 dimensions/inputs then there's the 11-Multiplexer (http://www.gene-expression-programming.com/webpapers/Ferreira-CS2001/Section6/SS5/SSS2.htm). You could make it a ten input problem by, e.g. defining the 11th bit as always 1 or 0.

Not sure if this is the kind of function you were lookign for though.

[stepleton.livejournal.com]

Are we to assume that it already works with something basic like a mixture of truncated 10-D Gaussians, and that you want something harder now?

[gustavolacerda.livejournal.com]

no.

I was going to do a mixture of truncated 10-D Gaussians... but it turned out to be more than a few minutes to program (my Matlab is surprisingly rusty after 3 months).

[gustavolacerda.livejournal.com]

This is interesting...

I wasn't expecting any "kind of function".


<< a random sampling of the 2048 combinations was used each generation as the fitness cases for evaluating fitness. >>

My objective should be deterministic. I suppose I could uniformly average its performance on the 2048 possible inputs. But this probably gets too expensive.

[gustavolacerda.livejournal.com]

but for the sake of honour, I must give this another try!

[the-locster.livejournal.com]

Yeh the performance is poor and random sampling doesn't 'feel' nice. On a similar theme how about an inverted pyramid of 2 input XORs, with your 10 inputs at the top and a single output at the bottom.