ozanarkancan / dqn-agent Goto Github PK

Learn how to work on Arcade Learning Environment by answering these questions and explain on wiki:

• What are those and how do they work?
  • Agent
  • Environment
  • Interface (FIFO etc.)
• How do we program an agent?
  • Is there a language restriction?
  • How do we get inputs and policies of games
• What command-line arguments are we going to utilise?

Research about Reinforcement Learning, currently-used algorithms (Q-Learning, Temporal Difference Learning, SARSA etc.) and become able to list pros and cons of those algorithms.

Read the paper and answer those questions and document the results on wiki

• What is the exact definition of the problem?
• What are inputs and outputs (for each timestep vs. for each game)
• How the input stream is used in their representation?
• In which fashion and how do they train their agent?
• Which games are used in ALE and how can we get their ROMs?