Bug in Validation for Differential Training

darioizzo / geodesyNets

Experiments with artificial neural networks and geodesy

GNU General Public License v3.0

19 stars 3 forks source link

Bug in Validation for Differential Training #83

Closed gomezzz closed 3 years ago

gomezzz commented 3 years ago

In the runs using https://github.com/darioizzo/anns4geodesy/pull/82/commits/c9dbe0d927956def60be18dcacda33fe0fc903ae there was a problem with the validation using differential training (very large errors). We should check that for correctness.

darioizzo commented 3 years ago

Its not a (proper) bug I think. Its only that when the gravity prediction comes from a differentially learned density we must add to the one coming from the reference model.

acc = acc_differential + acc_referencemodel

I think right now we only consider the first term