Comment by wholehog

See my comment above - the Nature authors already did this, and tried a huge hyperparameter sweep for SA, and RL still won.

See appendix of the Nature article: rdcu.be/cmedX