This run used a misconfigured dataset, where two variables had NaNs that were in non-masked region. Instead of erroring, the evaluation ran and had no error. However, prediction files are NaNs everywhere starting the very first step. It feels more natural that the emulator throws an error whenever any variable is NaNs everywhere.
This run used a misconfigured dataset, where two variables had NaNs that were in non-masked region. Instead of erroring, the evaluation ran and had no error. However, prediction files are NaNs everywhere starting the very first step. It feels more natural that the emulator throws an error whenever any variable is NaNs everywhere.