Photometric vs. ground-truth redshift for deterministic and probabilistic models trained on different approaches to mixing of GalaxiesML and TransferZ. The models are evaluated on a 28,000 sample of galaxies from GalaxiesML (top) and 11,000 sample of galaxies from TransferZ (bottom). The NN-Combo and BNN-Combo are models trained on a composite dataset approach while NN-TL and BNN-TL are trained on transfer learning from a base model trained on TransferZ and fine-tuned on GalaxiesML. The solid black line is a one-to-one line and the dashed black lines correspond to outliers with z_pred > ± 0.15(1 + z_true). The color scale indicates the density of data points (log scale). The deterministic models show less scatter and lower bias on both datasets than probabilistic models, with BNN-TL showing the most scatter.