Netlogo-pong

Simple netlogo implementation of a learning agent playing pong.

You can read a survey on "Learning Agents for Pong" here.

Folders

├── pong_exp                                # Experiments folder
├── survey                                  # LaTeX survey
|── Learning_Agents_for_Pong__A_Survey.pdf  # PDF survey
|── quality.csv                             # A quality matrix
└── pong.nlogo                              # Actual code

PLAY

If you want to see how an agent plays pong against a Scripted AI just do SETUP, LOAD (optional) and then PLAY. This way the agent is not learning but simply playing the best action based on the quality matrix (quality.csv). If you don't press LOAD the agent operates with an empty quality matrix (zeros only). In the pong_exp folder you can find quality matrices (quality_20k_episodes_rnd-prob-0.1.csv, etc.). By renaming them to quality.csv and putting them in the directory where the netlogo code is (pong.nlogo) you can see how the agent plays.

LEARN

You can generate your own quality matrix by doing SETUP, LOAD (optional), select the algorithm between SARSA and Q-Learning and then press LEARN. This way the agent is learning and generating a quality matrix (quality.csv). If you don't press LOAD the learning starts with an empty quality matrix (zeros only).

CONFIGURATION

Algorithm

You can choose whether the learning algorithm will be Q-Learning or SARSA.

State-type

You can choose whether, during learning, the opponent's state, that is the coordinate x, is known to the agent or not.

Reward-type

You can choose between two types of rewards. The basic reward is a reward that gives +1 in case of victory and -1 in case of defeat. The reward distance is based the basic reward but it also takes into accout the distance of the ball bounce from the center of the paddle.