Have you ever run code for several days only to get an error.
Serious question for structural people
-
No. I would never let it get to that point. Always test an application on a tiny test dataset first to iron out issues.
This is one right answer. Another option is: set your convergence criterion to 1e-2 just to make sure everything is running as you expect after the estimation is “done”.
Once it’s all good set it tighter.
-
No. I would never let it get to that point. Always test an application on a tiny test dataset first to iron out issues.
This is one right answer. Another option is: set your convergence criterion to 1e-2 just to make sure everything is running as you expect after the estimation is “done”.
Once it’s all good set it tighter.Setting convergence criterion is one the most arbitrary method. By the nature of variable’s scale, you can get arbitrary small or large criterion value. Setting 1e-5 can well mean your program takes 5 year to stop or stop in one iteration. It is more reliable to set the number of iterations.
-
No. I would never let it get to that point. Always test an application on a tiny test dataset first to iron out issues.
This is one right answer. Another option is: set your convergence criterion to 1e-2 just to make sure everything is running as you expect after the estimation is “done”.
Once it’s all good set it tighter.Setting convergence criterion is one the most arbitrary method. By the nature of variable’s scale, you can get arbitrary small or large criterion value. Setting 1e-5 can well mean your program takes 5 year to stop or stop in one iteration. It is more reliable to set the number of iterations.
better idea: set reasonably tight convergence criterion but limited the max number of iterations.