To reproduce results:

• Clone repository and get into it

git clone https://github.com/under-control/rl-task.git
cd rl-task

• Build docker:

docker build -t rl-task .

• Run docker with default parameters - 'FrozenLake8x8-v1' env for 100000 total learning timesteps:

docker run -it --rm  --network host --ipc=host --name rl-rask --mount src=$(pwd),target=/root/code/,type=bind rl-task bash -c "python /root/code/main.py"

Note: if you are on windows run docker commands on PowerShell (in order to use $(pwd) instead of typing your path)

Other options:

Run docker with parameters for 'CarRacing-v0' env with 1000000 total learning timesteps :

docker run -it --rm --network host --ipc=host --name rl-rask --mount src=$(pwd),target=/root/code/,type=bind rl-task bash -c "python /root/code/main.py -e 'CarRacing-v0' -s 1000000"

You can also get inside docker:

docker run -it --rm --network host --ipc=host --name rl-rask --mount src=$(pwd),target=/root/code/,type=bind rl-task bash

And make experiments while being inside:

python /root/code/main.py -e 'CarRacing-v0' -s 1000000

See results on tensorboard (examples folder contains run on default parameters)

tensorboard --logdir exp --port 6006