aaronspring
commented
2 years ago - save model as
.tf(hopefully not too large) - training (and eventually test) period predictions as
netcdfwith many parameters as attrs:- git commit
- timestamp
- git branch
- preprocessing details: number of EOFs, variables, hyperparameters of model build and train
Diagnostic function to fetch previous results and plot score easily would be nice
Some suggestions how to make our model reproducible: please add