Deep Reinforcement Learning Example that will build on Julia 1.0.1?


#1

Hi, Can someone recommend a good Deep Reinforcement Learning example code that will build on Julia 1.0.1. I am looking for Actor-Critic, policy gradient or even deep q learning example that works so I can then start doing fun stuff knowing everything builds. I have tried the following without luck.


(Gym won’t build)

(mountain_car.jl example fails)


#2

You could try running them with Julia 0.7 to see what deprecations need fixing, then fix those so they run on 1.0. Then make a Pull Request!


#3

I just updated CarloLucibello/DeepRLexamples.jl
for julia 1.0, please try it out.

You should also check out https://github.com/JuliaReinforcementLearning/ReinforcementLearning.jl,
I’ve never used it but seems a really nice package.


#4

Carlo, Thanks. I have actor_critic_pong.jp working! And, thank you for helping to make the Julia community great!


#5

thanks! wish I could do much more for the community. Fortunately is plenty of brilliant people around here.

little caveat: the cpu implementation of convolutions in Knet is rather slow, you should experiment much better performances running the code on a gpu


#6

Thanks Carlo. I just use the “Knet Array” type to get GPU to work right? I do have a fast GPU set up with NVIDIA.


#7

yep, if you are using KnetArray the deep net is running on the gpu. Have fun!