Home

rl

rl is a C++ reinforcement learning library, hence the name. It started as a project after taking an AI course with Richard Sutton, co-author of the popular Reinforcement Learning book. This wiki should guide you on how to use this library.

algorithms

• Discrete:
  • Sarsa
  • Q-learning
  • Dyna-Q
  • Dyna-Q with Prioritised Sweeping
  • Dyna-Q with Eligibility Traces
• Continuous:
  • Sarsa
  • Q-learning
  • Sarsa with Eligibility Traces
  • Q-learning with Eligibility Traces

This is of course not complete. For instance, in the near future, we will introduce the concept of TD-lambda, thus a developer can choose a pure TD to Monto-Carlo full backup.

Note-worthy features

• Allows for massive state-action space by utilising cassandradb. Thus one can represent their state-action space in multiple db server.
• Represent both supervised and unsupervised learning.

More features are to come and I can't wait to deliver them. If you have request feel free to request in the issue page.