Similar to Fig. 14, but including the predicted weights, and without any selection of ⟨S/N⟩. For the top panels, the field coordinates (x, y) are used as input features to the ML, while for the bottom panels, the PSF shape parameters ε₁, ε₂, and σ are used. To obtain μ_i and c_i for each bin, an unweighted linear regression is performed on the weighted average ĝ_i against the per case, instead of performing a weighted least squares regression on all realizations of a bin. The later approach would result in an undesired weighting between different PSFs.