Sanghyeok Choi
Sanghyeok Choi
이거 구현해주세요 ㅠㅜ
#248 여기에도 나와있듯 center 가 들어가면 비공개로 전환되는 것 같아요. 저도 그랬습니다! 근데 center를 왜 금지어로 설정해놓으셨는지..ㅠ
### Changes - Merged the updated main branch - Added the reward augmentation with local search, which is one of the key components of DeepACO.
### Changes - Merge the main branch (c.f., #170) - Increase the performance further (now it's very close to the performance of the original implementation of DeepACO) - Add CVRP...
Unresolved comments will be dealt with in the following PR.
> Good! Will the same hold with local search too @hyeok9855 ? Yes, very likely.
Personally, I don't want to include torchrl, only to use the buffer. I will first try refactoring without it. Let's see.
Regarding prioritization, the current implementation of `PrioritizedReplayBuffer` uses the reward (=priority score) just to sort the samples in the buffer and then drop low-reward ones when the buffer is full....
Can anyone explain why we need `terminating_states` ([here](https://github.com/GFNOrg/torchgfn/blob/a59b5810f619bcfb6dbb83058c672937c9fc44f5/src/gfn/containers/replay_buffer.py#L42)) in the replay buffer? It isn't used anywhere else outside of the buffer. If it's not necessary, I will remove it when...
Thanks @saleml. I've just checked the flow matching loss. I don't have a strong opinion, but I wonder if we could improve in this way: Can we simply add a...