Better Ending Conditions #9
Description
CMAES actually has a slurry of fine stopping conditions: https://github.com/CMA-ES/c-cmaes/blob/master/src/cmaes.c which we should start thinking about.
- Create a StopCondition OPERATOR that can be used by our Algorithms
- Make our current algorithms use the StopCondition instead of their current default impls (usually just amount of iterations), and always set a good default.
- Implement at least the following stopping conditions:
- Enough Iterations reached
- Enough EVALUATIONS done
- Fitness is good enough (value defined by user, default I suggest something like 1.0E-15)
- Some implementations may be specific to an alg:
- When the Fitness Value doesn't change anymore (See performConsistencyUpdate in CMAES)
- When the Standard-Deviation has no effect anymore (see original CMAES) (NoEffectAxis: standard deviation 0.1*%7.2e in principal axis %d without effect)