All the modifications are on top of the baseline pacman API from http://inst.eecs.berkeley.edu/~cs188/fa18/projects.html
Use minicontest1 for Qlearning agents and minicontest5 for MultiActor-CentralizedCritic agents
| Dependancy | Install Command |
|---|---|
| Numpy | python -m pip install numpy |
| Pickle | python -m pip install pickle-mixin |
| Tkinter | sudo apt-get install python-tk |
To train the model use
python pacman.py -n 10010 -l layoutName -g DirectionalGhost -r -q
This will add record files under minicontest1/records/ or minicontest5/records/
| Options | Description |
|---|---|
| -n GAMES, --numGames=GAMES | the number of GAMES to play, GAMES-numTraining will be used for testing (numTraining is a hyperparamter in the classes of agents in myAgents.py) [Default: 1] |
| -l LAYOUT_FILE, --layout=LAYOUT_FILE | the LAYOUT_FILE from which to load the map layout [Options: In layouts/ directory, Default: test51] |
| -q, --quietTextGraphics | Generate minimal output and no graphics |
| -g TYPE, --ghosts=TYPE | the ghost agent TYPE in the ghostAgents module to use [Options: DirectionalGhost, RandomGhost, Default: RandomGhost] |
| -r, --recordActions | Writes game histories to a file (last 10 episodes will be recorded) |
Use Bash script script.sh to run all the records under the minicontest1/records/ or minicontest5/records/ directory
OR
Use the following python command
python pacman.py --replay=FILNAME -l layoutName
layoutName should be consistent with training layout
- To change parameters for agents change the parameters change paramters in
__init__function of the corresponding agent class in myAgent.py - To change the mini-batch size change
minibatch_sizevariable inpacman.py(line number: 694)