phawkwood would say.I just started up 50 replications each of the Evolutionary Computation in Java (ECJ) jobs that
scottharmon, hermes_imagod, and chriszhong prepared (adapting Steve Gustafson's code and mine) for our 
geckies project on Incremental Learning in Multi-Agent Systems. This project, described in a forthcoming GECCO-2004 poster paper and a separate late-breaking paper, compares reuse strategies: automatically-defined functions (ADFs) and incrementally staged learning from easy subtasks (ISLES)1.For a single run with 101 generations, population size 4000, this thing takes about 1 hours (3600 seconds) on
Procyon, a Sun Blade 1000 (750MHz SPARC).On
Lakota, Osage, Kiowa, and Pawnee, it takes barely 1100 seconds. I don't think I'm even using the 64-bit, 4 processors, or a significant chunk of the 4Gb of RAM per processor.1 Yesh, I came up with that acronym. Aren't I a clever clogs? YRAINW.
Edit, 11:15 CDT: Delenda Est Banazir
Hey,
thanatos_kalos, deire, and phawkwood - you gotta see this story from elainemc. This is what teunc would call *nirglive* material...--
Banazir
(I'm
banazir, and I'm a HyperProf) 