We already assign :LN_BOBYQA as the default gradient-free optimizer when p == 1 (because :LN_NEWUOA punts on p == 1 problems). I have noticed that :LN_BOBYQA without any bounds enforced seems to converge in fewer iterations than :LN_NEWUOA when p is small (2 or 3). Should we change the threshold for the :LN_BOBYQA choice to say 3?
We already assign :LN_BOBYQA as the default gradient-free optimizer when p == 1 (because :LN_NEWUOA punts on p == 1 problems). I have noticed that :LN_BOBYQA without any bounds enforced seems to converge in fewer iterations than :LN_NEWUOA when p is small (2 or 3). Should we change the threshold for the :LN_BOBYQA choice to say 3?