cavis/rl4j
Chris Bamford 1a35ebec2e
RL4J: Add Backwardly Compatible Builder patterns (#326)
* Starting to switch configs of RL algorithms to use more fluent builder patterns. Many parameter choices in different algorithms default to SOTA and only be changed in specific cases

Signed-off-by: Bam4d <chris.bam4d@gmail.com>

* remove personal gpu-build file

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

* refactored out configurations so they are heirarchical and re-usable, this is a step towards having a plug-and-play framework for different algorithms

* backwardly compatible configurations

* adding documentation to new configuration classes

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

* private access modifiers are better suited here

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

* RL4j does not compile without java 8 due to previous updates

fixing null pointers when listener arrays are empty

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

* fixing copyright headers

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

* uncomment logging line

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

* fixing default value for learningUpdateFrequency

fixing test failure due to #352

Signed-off-by: Bam4d <chrisbam4d@gmail.com>

Co-authored-by: Bam4d <chris.bam4d@gmail.com>
2020-04-06 12:36:12 +09:00
..
contrib Eclipse Migration Initial Commit 2019-06-06 15:21:15 +03:00
rl4j-ale RL4J: Replace gym-java-client with JavaCPP (#8595) 2020-01-20 17:13:57 +09:00
rl4j-api RL4J: Replace gym-java-client with JavaCPP (#8595) 2020-01-20 17:13:57 +09:00
rl4j-core RL4J: Add Backwardly Compatible Builder patterns (#326) 2020-04-06 12:36:12 +09:00
rl4j-doom RL4J: Replace gym-java-client with JavaCPP (#8595) 2020-01-20 17:13:57 +09:00
rl4j-gym RL4J: Use Py_AddPath() instead of Py_SetPath() in GymEnv (issue #8688) 2020-02-19 00:31:12 +09:00
rl4j-malmo RL4J: Add Backwardly Compatible Builder patterns (#326) 2020-04-06 12:36:12 +09:00
LICENSE.txt Eclipse Migration Initial Commit 2019-06-06 15:21:15 +03:00
README.md RL4J: Replace gym-java-client with JavaCPP (#8595) 2020-01-20 17:13:57 +09:00
cartpole.gif Eclipse Migration Initial Commit 2019-06-06 15:21:15 +03:00
doom.gif Eclipse Migration Initial Commit 2019-06-06 15:21:15 +03:00
malmo.gif Eclipse Migration Initial Commit 2019-06-06 15:21:15 +03:00
pom.xml RL4J: Replace gym-java-client with JavaCPP (#8595) 2020-01-20 17:13:57 +09:00
scoregraph.png Eclipse Migration Initial Commit 2019-06-06 15:21:15 +03:00

README.md

RL4J: Reinforcement Learning for Java

RL4J is a reinforcement learning framework integrated with deeplearning4j and released under an Apache 2.0 open-source license. By contributing code to this repository, you agree to make your contribution available under an Apache 2.0 license.

  • DQN (Deep Q Learning with double DQN)
  • Async RL (A3C, Async NStepQlearning)

Both for Low-Dimensional (array of info) and high-dimensional (pixels) input.

DOOM

Cartpole

Here is a useful blog post I wrote to introduce you to reinforcement learning, DQN and Async RL:

Blog post

Examples

Cartpole example

Disclaimer

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

Quickstart

  • mvn install

Visualisation

webapp-rl4j

Quicktry cartpole:

  • run with this main

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"

Malmo (Minecraft)

Malmo

  • Download and unzip Malmo from here
  • export MALMO_HOME=YOURMALMO_FOLDER
  • export MALMO_XSD_PATH=$MALMO_HOME/Schemas
  • launch malmo per instructions
  • run with this main

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
  • Policy Gradient
  • Update rl4j-gym to make it compatible with pixels environments to play with Pong, Doom, etc ..