Performance validation vs. hydroDL and migrating 1.0 users

[leoglonz]

Summary

dPL Hydro models should yield identical results when trained/tested on the same GPU using hydroDL and gDM + hydroDL2. Furthermore, it should present as sufficiently similar to hydroDL such that users working with the 1.0 package can readily translate their work.

Actions

• Validation of training loss and test skill vs hydroDL for models dHBV1.0, dHBV1.1p, and dPRMS.
• Record training losses and metrics observed for future validation.
• Assessment of framework usability by hydroDL users with suggestions for improvement.

Considerations

• Testing will not happen until gDM and hydroDL2 are functionally complete.
• Ideally, tutorials should be ready by this time, and testing will only be a matter of running these tutorial scripts for dHBV1.0, dHBV1.1p, and dPRMS.
• Further validation against other dHydro models and implementation of any models without the tutorials would further ensure robustness.
• Low-demand testing results should be recorded for unit testing and any other validation.

Assignments: @DoaaAboelyazeed, @amirmoez98, @kelawson48, @kasProg, @Yalan-Song, @XuetingPu, @leoglonz