Having bounds for periodic parameters caused occasional fit failures. This should be more robust.
Particularly when combined with MLEFitterscipy.minimize would tend to try to take a point at the upper / lower bounds and could then get stuck trying to move away from the heuristic value (even if that was extremely close to the optimum). Removing these bounds and just clipping after the fit is conceptually cleaner and helps the optimizer.
Having bounds for periodic parameters caused occasional fit failures. This should be more robust.
Particularly when combined with
MLEFitter
scipy.minimize
would tend to try to take a point at the upper / lower bounds and could then get stuck trying to move away from the heuristic value (even if that was extremely close to the optimum). Removing these bounds and just clipping after the fit is conceptually cleaner and helps the optimizer.