You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Dec 7, 2022. It is now read-only.
Voigt et al. (2021) propose a new crossover operator. In short, a single individual from the population is used together with the best solution found so far. For every client $i$, if we have $i \rightarrow j$ in the best solution but $i \rightarrow k, j \ne k$ in the selected individual, then we place $i$ in a so-called removal pool. It could be seen as removing all "broken pairs" as in similar to the currently used diversity measure. When all pairs have been checked, a subset of clients from the removal pool are selected based on ALNS-type destroy operator criteria e.g., randomly, worst cost, etc. Those clients are removed from the individual and are then repaired again using greedy insert. Same idea in Voigt et al. (2022).
I tested this without success. Using SREX/BPX crossovers with the best solution leads to early convergence. It might be interesting to test variations where we do crossovers with best if (some condition), but given the importance of the dynamic problem, I think it's not worth to research this further.
Voigt et al. (2021) propose a new crossover operator. In short, a single individual from the population is used together with the best solution found so far. For every client$i$ , if we have $i \rightarrow j$ in the best solution but $i \rightarrow k, j \ne k$ in the selected individual, then we place $i$ in a so-called removal pool. It could be seen as removing all "broken pairs" as in similar to the currently used diversity measure. When all pairs have been checked, a subset of clients from the removal pool are selected based on ALNS-type destroy operator criteria e.g., randomly, worst cost, etc. Those clients are removed from the individual and are then repaired again using greedy insert. Same idea in Voigt et al. (2022).
Originally posted by @leonlan in https://github.com/N-Wouda/Euro-NeurIPS-2022/discussions/54#discussioncomment-3401551
Can we implement this ourselves?
The text was updated successfully, but these errors were encountered: