The notebook starts with a lengthy exploration of the optimiser behaviour, which is more about designing the problems than anything else
CMAES is wrapped, so can question the utility of testing it (although would still test controller)
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
Few things to note: