ajithalbus / policygradient

Implementation of Policy Gradient Learning algorithm on a 2D test-bed.

PolicyGradient

Implementation of Policy Gradient Learning algorithm on a 2D test-bed.

Requirements : gym,numpy,argparse,matplotlib

usage :

usage: rollout.py [-h] [--alpha ALPHA] [--gamma GAMMA] [--episodes EPISODES]
              [--env ENV] [--render] --batchSize BATCHSIZE

ALPHA : learning rate [Default:0.5]
GAMMA : Discount [Default:0.9]
--render : to render the game for last 10% of the episodes
ALGO : Q/SARSA/SARSAlam
ENV : [chakra/visham]

Example : To run a game with default settings on grid-A, to render the enviroment and show learnt policy at the end

python rollout.py --episodes 2000 --render --env chakra

Reference : Reinforcement Learning: An Introduction Book by Andrew Barto and Richard S. Sutton

Credits : The TAs of CS6700 Spring 2018 IITM for the Gym-Environment.