Source (GitHub)

  • Implemented Deep Q-Network (DQN) and its successor Double DQN for playing Atari Breakout using OpenAI - Gym
  • Developed a recurrent state mechanism to encode and maintain a history of actions and observations
  • Managed hyperparameters through “config.py” to achieve a mean score of 10 within computational limitations (around 2000 episodes)
  • Monitored and debugged progress using expected rewards versus the number of episodes as a benchmark
  • Demonstrated problem-solving and experimentation skills within the constraints outlined in the assignment
  • Successfully contributed to achieving the target evaluation score of 10 for either “agent.py” or - “agent_double.py”