Saturday, January 21, 2012

Feedback 21-01-2012


Work
  • Made changes according to comments from last meeting to thesis plan
  • Read several more papers
  • Research about available frameworks
    • RL Glue; supports Java, C/C++, Matlab, Python, Lisp
    • Maja Machine Learning Framework; Python
    • Reinforcement Learning Toolbox; C++
    • Several small environments (mostly in Matlab or Python)
  • Personal preference for RL GLue
    • Supports my preferred language; Java
    • Possibly best support of the framework options above
    • Multi-platform / -language
    • Used in the last RL Competitions
    • Online library with environments, agents and experiments
  • Installed RL Glue
    • Created Netbeans Java project sourcing all neccesary libraries
    • Created helper class to start up RL Viz, RL Glue, agent, environment and experiment
    • Changed build.xml to automatically build everything and run (in one click)
    • Downloaded RandomAgent, CartPole and MountainCar from rl-library for testing purposes
    • Created new environment; one shot - Six-Hump Camel Back 
Plans
  • Investigate RL Glue environment more
  • Finish Six Hump Camel Back One shot environment (visualization)
  • Find environments taking continuous actions (not in rl-library; perhaps from last year's RL competitions)
  • Change Cart Pole environment to take continuous actions

No comments:

Post a Comment