Currently, all clusters produce a track (it just iterates until the max_num_iterations and picks the best one).
I think that testing convergence within the Nelder-Mead sd_tolerance doesn't make a lot of sense. Maybe there should be some better way to just say "this best track is trash".
Maybe compare how far off the initial guess it is (?).
Not completely sure yet.
Currently, all clusters produce a track (it just iterates until the
max_num_iterationsand picks the best one).I think that testing convergence within the Nelder-Mead
sd_tolerancedoesn't make a lot of sense. Maybe there should be some better way to just say "this best track is trash".Maybe compare how far off the initial guess it is (?). Not completely sure yet.