Escaping the State of Nature

Thesis Project 2018-2019 Code: Escaping the State of Nature

Game

Players capture territory in a gridlike environment and can choose to cultivate their land or invade other players' territory for some bonus reward, but also accrue penalties when being invaded themselves.

Install Instructions

1. Download the primary dependencies with

pip install -r requirements.txt

2. Set the desired game parameters in params.py and then run trials with

python state_of_nature.py Q Q -3 Q -4 Q -s 4 -t 3 -hp -w 

You'll need to login with a plot.ly account to write graphs

3. Find matrix game Q-networks in assets/ with

python play_matrix_game.py

Initial Findings:

Q-Learning in naive settings

Q-Learning vs Q-Learning Average Score

Q-Learning vs HQ-Learning Collective Score

Q-Learning vs HQ-Learning Action Breakdown

Comparative Collective Invasions

Comparative Successful Defers

Social Dilemma Matrix Game

	C	D
C	(R = 0.459, R = 0.459)	(S = 0.426, T = 0.446)
D	(T = 0.446, S = 0.426)	(P = 0.455, P = 0.455)

Criteria:

1. R > P
2. R > S
3. 2R > T + S
4. T > R or P > S

Fear = P - S (+0.029), Greed = T - R (-0.013)

If both Fear and Greed exist in the Civilization Game social dilemma, it represents a Prisoner's Dilemma matrix game. If just Fear, then a Stag Hunt.