Posterior probability distribution estimated trained on the D2 dataset (left) and training on the D4 dataset (right) (Table 2). These datasets both come with no dn/dz selection but differ in the choice of fixing the γ parameters. Four MDNs are trained per dataset each on a different survey area. These results are computed for the same target testing XOD, always of the appropriate survey size, that was produced having all 8 parameters free and with the dn/dz selection. We can again observe that using the dn/dz selection and fixing the γ parameters improves the accuracy of the NDE’s cosmological predictions.