[ \text{CWDR}(i,j) = \frac{d_{ij}}{ \max\left( \epsilon,\ \frac{Q_{rem}}{q_j} \right) } ]
Below is a you can engineer for a CTSP solution (e.g., when training a neural network or constructing a heuristic): Feature: Capacity-Weighted Distance Ratio (CWDR) Definition For each candidate customer node ( j ) relative to current node ( i ) and remaining vehicle capacity ( Q_{rem} ): j) = \frac{d_{ij}}{ \max\left( \epsilon
At decision time, concatenate RCLR with node embeddings. This helps the model learn — a critical CTSP decision absent in TSP. If you meant a different CTSP (e.g., in biology or signal processing), let me know and I'll tailor the feature suggestion accordingly. in biology or signal processing)