smarthi / rl4j

Reinforcement Learning for the JVM

RL4J

Reinforcement learning framework integrated with deeplearning4j.

GIFS are coming

temporary example repo

Disclaimer

This is a tech preview and distributed as is. Comments are welcome on our gitter channel: gitter

Quickstart

• mvn install -pl rl4j-api
• [if you want rl4j-gym too] Download and mvn install: gym-java-client
• mvn install

Quicktry cartpole:

• Install gym-http-api.
• launch http api server.
• run with this gist

Doom

Doom is not ready yet but you can make it work if you feel adventurous with some additional steps:

• You will need vizdoom, compile the native lib and move it into the root of your project in a folder
• export MAVEN_OPTS=-Djava.library.path=THEFOLDEROFTHELIB
• mvn compile exec:java -Dexec.mainClass="YOURMAINCLASS"

WIP

• Documentation
• Serialization/Deserialization (load save)
• Compression of pixels in order to store 1M state in a reasonnable amount of memory
• Async learning: A3C and nstep learning (requires some missing features from dl4j (calc and apply gradients)).

Author

Ruben Fiszel

Proposed contribution area:

• Continuous control