Review for NeurIPS paper: Hard Shape-Constrained Kernel Machines

Additional Feedback: Do you have any comments on how your hard shape constraint formulation affects overlapping quantiles versus the soft constraint (PDCD)? It would be more informative to display Figure 1 alongside plots from alternative methods. Is it reasonable to attribute the comments on l304 regarding non-crossing to the additional regularizing properties of the concavity constraint? Regarding Table 1, it seems SOC performs _worse_ than PDCD on 5/9 datasets. This opens the question of when and where are hard constraints more beneficial over soft constraints.