JuliaGymDemo

Train and Play Gym with Julia v0.6 + Gym.jl(revised) + Knet.jl

Prerequisites

What things you need to install the software and how to install them

Target Julia and Packages

Julia v0.6.x
- Gym.jl
- ArgParse.jl
- JLD.jl
- Knet.jl

Simple Installation

Run the codes on Julia REPL:

julia> Pkg.clone("https://github.com/antimon2/Gym.jl.git")
julia> Pkg.checkout("Gym", "mln_ngy")
julia> Pkg.build("Gym")
julia> Pkg.add("ArgParse")
julia> Pkg.add("JLD")
julia> Pkg.add("Knet")

Custom Installation with installed `Gym`

If you have already installed Gym on your own Python (for instance, on the path s.t. /path/to/user_home/.pyenv/versions/3.6.x/bin/python):

julia> ENV["PYTHON"] = "/path/to/user_home/.pyenv/versions/3.6.x/bin/python"
julia> Pkg.clone("https://github.com/antimon2/PyCall.jl.git")
julia> Pkg.checkout("PyCall", "mln_ngy")
julia> Pkg.clone("https://github.com/antimon2/Gym.jl.git")
julia> Pkg.checkout("Gym", "mln_ngy")
julia> Pkg.build("Gym")
julia> Pkg.add("ArgParse")
julia> Pkg.add("JLD")
julia> Pkg.add("Knet")

Getting Started

Clone this repository, run julia cartpole_sample.jl:

$ git clone https://github.com/antimon2/JuliaGymDemo
$ cd JuliaGymDemo
$ julia julia cartpole_sample.jl
WARN: gym.spaces.Box autodetected dtype as <class 'numpy.float32'>. Please provide explicit dtype.
episode 1 total Rewards: 21.0
episode 2 total Rewards: 13.0
episode 3 total Rewards: 48.0
episode 4 total Rewards: 16.0
episode 5 total Rewards: 15.0
episode 6 total Rewards: 14.0
episode 7 total Rewards: 46.0
episode 8 total Rewards: 12.0
episode 9 total Rewards: 14.0
episode 10 total Rewards: 33.0
$

Then you can see outputs like above and window like below:

for Trainings and Tests

cartpole_train.jl: Train the Gym Env CartPole-v0
- julia cartpole_train.jl -h for help.
cartpole_play.jl: Play the Gym Env CartPole-v0 with trained model saved by cartpole_train.jl
- julia cartpole_play.jl -h for help.

antimon2/JuliaGymDemo