NaNs in learning if PES file contains entries without energies

[adabbott]

If a PES file used to train a NN or GP has geometries without energies and a model is trained, one gets NaNs during training:

Hyperparameters:  {'layers': (16, 16), 'morse_transform': {'morse': False}, 'pip': {'pip': False}, 'scale_X': {'activation': 'tanh', 'scale_X': 'mm11'}, 'scale_y': 'mm01', 'lr': 0.5}
Epoch 1 Validation RMSE (cm-1):   nan
Epoch 2 Validation RMSE (cm-1):   nan
Epoch 3 Validation RMSE (cm-1):   nan
Epoch 4 Validation RMSE (cm-1):   nan
Epoch 5 Validation RMSE (cm-1):   nan
Epoch 6 Validation RMSE (cm-1):   nan
Epoch 7 Validation RMSE (cm-1):   nan
Epoch 8 Validation RMSE (cm-1):   nan
Epoch 9 Validation RMSE (cm-1):   nan
Epoch 10 Validation RMSE (cm-1):   nan
Epoch 11 Validation RMSE (cm-1):   nan