Why do you ignore the 0s?

We could not get a 100% correlation when comparing the ground truth to itself. So we filtered out the zeros.

After an analysis where we recalculated all the predictions keeping the zeros, the ranks are almost exactly the same.