from branch
This is a Reinforcement Learning framework with the purpose of experimenting with RL and building and training different agents.
The status of the automatic tests can be viewed here
Test progress and coverage is visualized here
Docker images used for development can be found here
To get start with development, run docker-compose up -d to start the docker environment that supports the python packages needed.
To access the environment with a terminal, run bash attatch.sh
you can edit the code in the local repository on your machine and test the code immediately in the docker image.
You can log the training using a tensorboard object that logs the training result and logs by default in a folder called runs in files with names created from a timestamp.
To start tensorboard:
tensorboard --logdir src/agents/agent_storage
This will start tensorboard on localhost:6006 from the runs folder
Docker has the purpose of creating an environment the different parts of the application can run on. See the docker docs
To enter a container that is running, type
docker exec -it [container-id] bash
Setting up a docker volume, i.e. a dynamic connection between a folder on the local host and the container, can be problematic on windows. If docker is unable to establish a volume connection due to shared drivers try the following fix:
Omniboard and mongo DB must be run and also be able to talk with each other. This can be obtained by starting them on the same docker network. First, create a new docker network or use an existing network
docker network create omniboard-network
We now have a network on which to run the docker containers. The mongodb and omniboard container should use the same docker network
docker run --rm --name mongo-container --net omniboard-network -d mongo
Then run the omniboard network
docker run --rm -d -p 9000:9000 --name omniboard --net=omniboard-network vivekratnavel/omniboard -m MONGODB_CONTAINER:27017:sacred
To start up the RL environment with a jupyter notebook running, write:
docker run --rm -it -v pwd:/notebooks -p 8888:8888 justheuristic/practical_rl
Go to localhost:8888 and insert the token from the console to log in. A RL environment image has been made for this projects and can be run by
docker run --rm -it -p 8888:8888 fabiansd/rl-env bash
You will then start up a linux container with all the necessary libraries installed. Here you can run python scripts and linux commands, and also start jupyter by typing
sh /RoboSchool/src/run_jyputer.sh
The frontend application is implemented with a Flask app and docker image built from /Roboschool with
docker build -f app/Dockerfile -t fabiansd/roboschool-app .
The application can be started up by running:
docker run --rm -d -p 9999:9999 fabiansd/roboschool-app
Then go to localhost:9090 to see the frontend application
Docker-compose is used to start all the docker files for the appliaction and enables you to run a single command to start up everything. I short: it summarizes all the commands above into one file and executes them such that all you have to do is run the docker-compose file and you're good to go.
To run application with docker-compose:
docker-compose up -d
The running containers can now be viewed with
docker ps
To enter the development environment, attach to the src container by copying its ID in the container list and type
docker exec -it [container-id] bash
To close the containers, i.e. the application, simply type
docker-compose down