search

IHCantabria / NEOPRENE

Neyman-Scott Process Rainfall Emulator library
GNU General Public License v3.0
13 stars 5 forks source link

Check validation function #6

Closed manueldeljesus closed 2 years ago

manueldeljesus commented 2 years ago

The validation functions seems to be computing the simulated statistics at a different temporal scale than the one used for fitting, giving the impression that the fits are worse than they really are. We should check why.

JavierDiezSierra commented 2 years ago

The problem comes from the hyperparameter Seasonality_type. In the calibration and the simulation yamls this hyperparameter is now set as "seasonal" and "monthly", respectively. If we want to compare the same observed, fitted and simulated statistics this must be the same.

@navass11, would you please re-run the NSRP notebook with the same argument in both yaml files. We must also modify the notebooks to omit the name of the locations (Bilbao, Sevilleta, ...)

navass11 commented 2 years ago

@JavierDiezSierra and @manueldeljesus I have already made the changes and re-run the code so that the figure is correct.

JavierDiezSierra commented 2 years ago

@navass11 I think it would be better to set verbose = False during the calibration to avoid to include unnecessary lines in the notebooks.

navass11 commented 2 years ago

@manueldeljesus @JavierDiezSierra The notebooks have been updated with the recommendations that were proposed