 |
 |
David, sorry for the delay -- this got spamtrapped somewhow. On Aug 4, 2007, at 11:45 PM, David Oranchak wrote: > Inspecting ec.steadystate.SteadyStateEvolutionState, I noticed this > line: > > // CHECKPOINTING > if (checkpoint && generation%checkpointModulo == 0) > > I reasoned that "generation" is only incremented during generation > boundaries, which for steady state occurs when the number of > evaluations reaches some multiple of the population size. Thus, > the checkpointing will continue unnecessarily until the next > generation boundary is reached. > > Here is my bandaid: > > // CHECKPOINTING > if (checkpoint && generation%checkpointModulo == 0 && > evaluations%checkpointModulo == 0) > > Does this look like a valid bug and fix? I am reluctant to check > anything into CVS without review. This is definitely a bug, but I'm not sure if the bugfix will work right; rather, it should probably be modding evaluations by the population size or something. But I would hold off on a fix -- Keith and I have an experimental brand new SteadyStateEvolutionState which promises to allow for massively parallel steady state evolution in a very cool way. It should be a drop-in replacement for the existing one and you could test that. Sean