rename the folderDefs_example.py file by removing "_example" and adjust with your path. (do not commit the resulting folderDefs.py as it is specific to your system and should not be propagated to mine...)
To create the tf records, run eg:
python ./dataLoad.py --input_names='TAP,QAP,SHFLX,LHFLX,PS' --output_names='SPDT,SPDQ' --convert_units=true --convo=true
Then, to run experiments:
python ./main.py --input_names=TAP,QAP,,SHFLX,LHFLX,PS' --output_names=SPDT,SPDQ --convert_units=true --convo=true --localConvo=true --hidden=64,64,64,64,64,64 --batch_size=128 --optim=adam --lr=1e-3 --frac_train=0.8 --log_step=500 --epoch=10
for non convolutional you must generate the records also (it will make separate ones).