It would be nice to fail early in certain cases.
Pro
- For example, a flat line with noise or a straight line with noise can both lead to maximum iterations being hit (if the noise is just right).
- Stats tests for "is this linear" are possible, if we have a secondary estimate of sample standard deviation s
Con
- For more complex models, or just very unlucky noise samples, it can still fail
- The method can only do what you ask it.
It would be nice to fail early in certain cases.
Pro
Con