-
Notifications
You must be signed in to change notification settings - Fork 134
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
revisit underflow trapping approach and exp_argmax values #170
Comments
Try uncommenting lines 563, 564 in icedrv_step.F90. This could fix a number of problems, e.g. issues #48 , #167 . |
I search the code (cicecore and icepack) for underflow instances, and did not find others besides the ones in evp and eap and the ones that were discussed and/or fixed here:
I also looked at all of the places where puny (10^-11) is used. These are mostly to prevent divide-by-zero, for masking purposes, or for certain physical states (e.g. snow depth disappears during melting, or the sun being below the horizon). These all seem fine to me. I propose the following:
|
Currently, the underflow trapping is not entirely consistent, which maybe is fine. But it was done somewhat adhoc. Review and modify if appropriate. For instance, maybe we should compute ln(tiny) to define the exp_argmax instead of choosing it arbitrarily. Or maybe exp_argmax should be more uniform across the code. Or maybe there are other approaches.
The text was updated successfully, but these errors were encountered: