search

davidhershey / feudal_networks

An implementation of FeUdal Networks for Hierarchical Reinforcement Learning as published : https://arxiv.org/abs/1703.01161
MIT License
178 stars 46 forks source link

Does code achieve benchmarks #7

Open awearytraveler opened 6 years ago

awearytraveler commented 6 years ago

Does this code achieve the benchmarks given in the paper? I modified this to work on my system, but it doesn't converge after running it for a few days.

davidhershey commented 6 years ago

Unfortunately we were never able to get Feudal Networks to converge successfully. I've talked to some other researchers that have had difficulty reproducing the paper as well. Its likely that there are a few implementation differences as the paper doesn't fully describe the architecture.

CrazySssst commented 6 years ago

I met same problem, loss doesn't decrease.

lucasliunju commented 6 years ago

Hello! I'm very sorry to trouble you! I want to ask whether the mode (tmux, child) make a great influence in training. Because I have trouble in tmux and I don't know why.