The CMAES notebook starts with a visual exploration of the optimiser behaviour, which is also useful for designing the problems
CMAES is wrapped, so can question the utility of testing it - but likely still a good idea to see if our wrapping and controller etc. are OK
For the future changepoint testing, it might make most sense to pick a number of iterations where we're generally about 2/3s of the way towards convergence, so that we get a distribution to compare? Alternatively have the output be e.g. "error at 50" and "error at 100", and do changepoint testing on both? Not the major concern of this PR, but something to think about for the future :D
Part of #20
Few things to note: