PacmanDQN

Deep Reinforcement Learning in Pac-man

Usage

Play 4000 games (3000 training, 1000 testing)

$ python3 pacman.py -p PacmanDQN -n 4000 -x 3000 -l smallGrid

Parameters

Parameters can be found in the paramters dictionary in pacmanDQN_Agents.py.__ __ Models are saved as "checkpoint" files in the \saves directory.__ Load and save filenames can be set using the load_file and save_file parameters.__ __ Episodes before training starts: train_start__ Size of replay memory batch size: batch_size__ Amount of experience tuples in replay memory: mem_size__ Discount rate (gamma value): discount__ Learning rate: lr__ RMS Prop decay rate: rms_decay__ RMS Prop epsilon value: rms_eps__ __ Exploration/Exploitation (ε-greedy):__ Epsilon start value: eps__ Epsilon final value: eps_final__ Number of steps between start and final epsilon value (linear): eps_step__

Requirements

• python==3.5.1
• tensorflow==0.8rc

Acknoledgemenets

DQN Framework by (made for ATARI / Arcade Learning Environment)

• deepQN_tensorflow (https://github.com/mrkulk/deepQN_tensorflow)

Pac-man implementation by UC Berkeley:

• The Pac-man Projects - UC Berkeley (http://ai.berkeley.edu/project_overview.html)