search

ai4co / rl4co

A PyTorch library for all things Reinforcement Learning (RL) for Combinatorial Optimization (CO)
https://rl4.co
MIT License
381 stars 70 forks source link

[BugFix] Fix the performance issue of DeepACO #170

Closed Furffico closed 3 months ago

Furffico commented 4 months ago

Description

This pull request fixes the performance issue of the DeepACO implementation, and makes some tiny modifications to enhance the performance.

Furffico commented 4 months ago

Made some modifications following the discussions above. BTW, rl4co/models/zoo/deepaco/decoder.py seems redundant. Is it ok to delete this file?

fedebotu commented 4 months ago

I guess so, we had a look at that too. Right @hyeok9855 ?

fedebotu commented 4 months ago

@Furffico did the above fixes make DeepACO match the performance in your paper?

Furffico commented 4 months ago

@Furffico did the above fixes make DeepACO match the performance in your paper?

Yes. With a parallelized AS and a better pheromone update function, it's faster and sometimes perform better than our original implementation.

fedebotu commented 4 months ago

Yes. With a parallelized AS and a better pheromone update function, it's faster and sometimes perform better than our original implementation.

Good! Will the same hold with local search too @hyeok9855 ?

hyeok9855 commented 3 months ago

Good! Will the same hold with local search too @hyeok9855 ?

Yes, very likely.

Furffico commented 3 months ago

I think there's no further modifications required in this PR. Shall we merge it now?