/Reinforcement-Learning

Understanding and working on Reinforcement Learning

Primary LanguageJupyter Notebook

Reinforcement-Learning

Understanding and working on Reinforcement Learning

About the folders:

  1. Multi-armed bandits

    Implementation of Multi-armed bandits algorithm for 10 armed test-bed using :

    1. greedy algorithm
    2. epsilon-greedy algorithm
    3. UCB

  2. Monte Carlo Simulation

Implementation of Monte Carlo first visit algorithm for Grid-world problem.

  1. Implementation of Q-learning on the game of Pacman
  • The code of Pacman game is downloaded from the website of UC Berkeley.
  • The implementation of Q learning algorithm is in the file qlearningagents.py.
  • To run the game, download all files including layout folder from the repo, maintaining the similar folder structure as on repository.
  • Open terminal/cmd, change the directory to the same directory where the pacman.py file exists.
  • Run below mentioned commands to see either agent playing the game or You.
    • python pacman.py - runs the game and can be played by You
    • python pacman.py --help - shows all the options available
    • Q-learning agent will play the game in different map layouts as mentioned at the end of the commands.

    python pacman.py -p QLearnAgent -x 2000 -n 2010 -l smallGrid
    python pacman.py -p QLearnAgent -x 2000 -n 2010 -l testClassic
    python pacman.py -p QLearnAgent -x 2000 -n 2010 -l trappedClassic
    python pacman.py -p QLearnAgent -x 2000 -n 2010 -l minimaxClassic
    python pacman.py -p QLearnAgent -x 2000 -n 2010 -l mediumGrid

Python version: 3.7.3