I am using optuna to maximize my criteria. More specificaly, I am using CMA-ES sampler with "ipop" strategy.
sampler = optuna.samplers.CmaEsSampler(restart_strategy="ipop",inc_popsize=2,)
As far as I understand it CMA-ES is supposed to restart with increasing population size when an optimum is reached. But in my case, it seems that there is no restart. I want to maximize my output, I reach almost my maximum at trial ~1/2k but nothing happen after that, CMAES is only doing exploitation, where I expect exploration.
Even if I do not clearly understand the restart conditions, I would expect at least one restart with that much trials.
(see paper http://www.cmap.polytechnique.fr/~nikolaus.hansen/cec2005ipopcmaes.pdf )
So my question is, why is optuna not restarting ?
I have seen that on optuna website:
I hope it is just a typo and this also work for local maximum.
Thanks in advance for your feed-backs.