Ms Pacman Documentation

Hyperparameter tracker.xlsx

Records on previously trained policies

Hyperparameter_Optimizer.ipynb

Use to find optimized hyperparameters for RL agent

Hyperparameter_Optimizer_sample.html

Sample output of 50 Optuna trials in Hyperparameter_Optimizer.ipynb

MS_Pacman_PG.ipynb

Use to train policy network to play Ms. Pac-Man

MS_Pacman_PG_sample.html

Sample output of 100 training episodes in MS_Pacman_PG.ipynb

MsPacManPG_optimized.pt

Policy network object of RL agent trained on Ms. Pac-Man

MsPacMan_study.db

Optuna study data after final changes to rewards system

visuals.py

Functions for visual representation of training/testing data

Archive

ms_pacman.py & policyGradient.py

Policy network based off of Github user ritesh-kanchi
This code was not used later on for this project.

Gifs

pac_man1.gif & pac_man2.gif

Visualization of training progress for our final version of policy network

Policies

Trained policy network objects

MsPacManPG_1.pt & MsPacManPG_2.pt

Created during development of improved policy
Not final versions of policy

MsPacManPG_optimized.pt

Final version of policy trained on 10000 episodes

Replay

Gameplay renderings of different trained policies

MsPacMan_beforeOptimizing_replay.mp4

Gameplay of policy network from Hugging Face Deep Reinforcement Learning course with minimal changes and 10000 episodes of training

MsPacMan_replay_0.mp4, MsPacMan_replay_1.mp4, MsPacMan_replay_2.mp4

Created during development of improved policy
Not gameplay from final versions of policy

MsPacMan_replay_3_best.mp4 & MsPacMan_replay_3_worst.mp4

Best and worst gameplay of final trained policy over 1000 episodes

Scores

Scores of each episode during training sessions

scores_afterOptimizing.csv

Training of final version of policy network over 10000 episodes

scores_beforeOptimizing.csv

Training of initial version of policy network from Hugging Face Deep Reinforcement Learning course with minimal changes over 10000 episodes

Studies

Optuna study data for finding the best hyperparameters for training Ms. Pac-Man RL agent

MsPacMan_study.db

Study ran after making final changes to reward system