Game Theory - Axelrod Tournament
This Python program simulates a tournament in the vein of the Axelrod Tournament, where different strategies compete against each other in the game of the Prisoner's Dilemma. Strategies include TitForTat and AlwaysDefect for example. This Tournament is an attempt to replicate and help understand the works of "Strategies for the Iterated Prisoner’s Dilemma" (Anagh Malik, 2020), and is heavily influenced by the Youtube channel Veritasium and their presentation "What The Prisoner's Dilemma Reveals About Life, The Universe, and Everything".
The key component to replicating the results presented in the Veritasium presentation is the random error generator. This is to simulate false positives/negatives to simulate real world situations. Veritasium use the example:
"On 26 September 1983, during the Cold War, the Soviet nuclear early warning system Oko reported the launch of one intercontinental ballistic missile with four more missiles behind it, from the United States. Petrov, suspecting a false alarm, decided to wait for a confirmation that never came."
This is a work in progress and created out of curiosity. The end goal is to simply replicate the results of the Axelrod Tournament and setup some simple python code for others to expand on if that should tickle one's fancy.
Description
Classes
GameRunner
The GameRunner class runs a specified number of games between two strategies and calculates the scores.
The Tournament class orchestrates multiple GameRunner sessions for various pairs of strategies, keeping an overall tally of scores.
The Strategy class is the base class for different strategies. It includes common methods like make_choice, set_choice and get_choice. It includes common attributes including name and init_choice.
Subclass examples: TitForTat is a strategy that cooperates initially and then mimics the opponent's last move.
AlwaysDefect is a strategy that always chooses to defect.
Example usage:
I simply run the prisoners_dilema.py script in a python shell to retrieve the output. Or something like PyCharm/VS-Code/Jupyter-Lab will enable you to see the output printed to the screen. The below shows the code that is tacked onto the end of the script. Update to modify the output.
from Strategies import *
# No random elements # Add more strategies as needed
strategies_non_random = [TitForTat, AlwaysDefect, GenerousTitForTat, ModalTFT, AlwaysCooperate,
Friedman, Graaskamp, Nydegger, DefectOnce, TitForTwoTats, WinStayLooseShift,
Benjo, Shubik, Downing]
# Introduced random elements
strategies_random = [TitForTat, AlwaysDefect, GenerousTitForTat, Joss, TidemanChieruzzi, Random,
AlwaysCooperate, Friedman, Graaskamp, Nydegger, DefectOnce, Downing,
TitForTwoTats, WinStayLooseShift, Benjo, Shubik, ModalTFT]
# All the strategies
strategies_all = [TitForTat, AlwaysDefect, AlwaysCooperate, GenerousTitForTat, Friedman,
Joss, Graaskamp, TidemanChieruzzi, Nydegger, TitForTwoTats, Random, Shubik,
WinStayLooseShift, Benjo, ModalTFT, ModalDefector, Downing]
# games = random.randint(200, 1000)
games = 2000
# Running games
tournament = Tournament(strategies_all, num_games_per_match=games, noise=True) # Mess with the parameters if you want.
overall_scores = tournament.round_robin()
sorted_scores = sorted(overall_scores.items(), key=lambda x: x[1], reverse=True)
print("")
print('*'*44)
print(" Overall Scores, sorted by highest ranking:")
print('*'*44)
print(f"Number of games: {games}\n")
for strategy, score in sorted_scores:
print(f"{strategy:>20}: {score}")
print('*'*44)
Example output
********************************************
Overall Scores, sorted by highest ranking:
********************************************
GenerousTitForTat: 4475
TitForTat: 4067
Friedman: 4064
AlwaysDefect: 3837
AlwaysCooperate: 3832
********************************************
********************************************
Overall Scores, sorted by highest ranking:
********************************************
Nydegger: 7852
AlwaysCooperate: 7201
GenerousTitForTat: 7157
TitForTat: 6692
Graaskamp: 6421
Friedman: 6074
AlwaysDefect: 5139
Joss: 5090
********************************************
********************************************
Overall Scores, sorted by highest ranking:
********************************************
Number of games: 303
TitForTwoTats: 15191
Nydegger: 14571
GenerousTitForTat: 14360
AlwaysCooperate: 13977
TitForTat: 13168
Random: 13130
Graaskamp: 12810
Joss: 11382
Friedman: 11051
AlwaysDefect: 10284
********************************************
********************************************
Overall Scores, sorted by highest ranking:
********************************************
Number of games: 2000
GenerousTitForTat: 74207
TitForTat: 72877
TitForTwoTats: 72112
Benjo: 71256
Random: 66892
ModalTFT: 65699
AlwaysCooperate: 64952
WinStayLooseShift: 64364
Nydegger: 62681
Graaskamp: 61749
Joss: 60123
Friedman: 59906
TidemanChieruzzi: 59730
Shubik: 59649
AlwaysDefect: 59537
ModalDefector: 59421
Downing: 58230
********************************************
References
Malik, A. (2020). Strategies for the Iterated Prisoner’s Dilemma. November 2020. Retrieved from https://arxiv.org/pdf/2111.11561.pdf
Muller, D. (2023, December 24). What The Prisoner's Dilemma Reveals About Life, The Universe, and Everything. Retrieved from https://www.youtube.com/watch?v=mScpHTIi-kM