Group Meeting - Tuesday 5th March

[vviers]

• [ ] How do we deal with particles getting out of bounds?

  • kill them and randomly assign new one?
  • let them be?
  • bounce?
  • different options?

• [ ] Convergence criterion?

  • Particle moves less than a certain distance for more than k iterations
  • Best global error has not changed for more than k iterations
  • both?

• [ ] ~The current parallel implementation is painfully slow. Pool() clusters are opened and closed at every iteration. Is there a way to keep it open?~

[vviers]

Maybe for apply_async --- > return a tupple (particle_id, fitness)