-
Notifications
You must be signed in to change notification settings - Fork 2
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
Pressure Level Scalings only applied in specific circumstances #7
Comments
What should be the behaviour if not specified? It seems that not doing anything is the appropriate response |
I think whether you scale the loss across pressure levels should be independent of whether you use the default loss scaling for the variable.
|
So, you want the pressure level scaling independant of loss:
_target_: anemoi.training.losses.mse.WeightedMSE
scalars: ['variable', 'pressure_level']
# or
scalars: ['pressure_level'] |
That's a cool functionality! But I'm not sure it currently works given https://github.com/ecmwf/anemoi-training/blob/develop/src/anemoi/training/train/forecaster.py#L303? |
It was a suggestion, yes, it won't currently work, and will require a refactor. |
What happened?
Pressure level scalings are only applied if the variable is specified in the variable loss scaling. This is because of this line https://github.com/ecmwf/anemoi-training/blob/develop/src/anemoi/training/train/forecaster.py#L303
What are the steps to reproduce the bug?
Specify for example
training:
pl:
t: 1
in the training config. You will then see that only the temperature is scaled by pressure level.
Version
0.3.1
Platform (OS and architecture)
Relevant log output
No response
Accompanying data
No response
Organisation
No response
The text was updated successfully, but these errors were encountered: