Main results of correcting z_photo-dependent residuals with the SDSS data and the network Net_P. Upper panels: spectroscopic redshift and r-band magnitude distributions for the SDSS training data and the test data. In addition to the original test sample, two other test samples are constructed that are biased toward medium redshifts (z_spec ~ 0.12) or high magnitudes (r ~ 17.5). Middle panels: mean residuals as a function of spectroscopic redshift for the baseline method and step 3 with the three test samples and the network Net_P. The results from step 4 are also plotted in light purple for comparison. Steps 1-3 are applied in advance of the calibration by step 4. Lower panels: mean residuals as a function of photometric redshift for the baseline method and step 4 with the three test samples and the network Net_P. The results from step 3 are also plotted in light red for comparison.