mirror of
https://github.com/taxicomics/PrisonersDilemma.git
synced 2025-03-19 22:16:05 +01:00
31 lines
1.2 KiB
Python
31 lines
1.2 KiB
Python
|
|
#basic setup
|
|
#The prisoner's dilemma is a game theory thought
|
|
#experiment that involves two rational agents, each
|
|
#of whom can COORPERATE for mutual benefit or betray
|
|
#their partner ("defect") for individual reward.
|
|
|
|
#imports
|
|
import helper_functions
|
|
import contenders
|
|
#vars
|
|
#keeping track of the individual games
|
|
games=[]
|
|
#keeping track of the games scores for later analysis
|
|
scores=[]
|
|
nr_of_rounds=30#int(input("How many should be played in a game?"))
|
|
nr_of_games=300#int(input("How many games should be played?"))
|
|
#XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
|
|
# Put YOUR functions down below
|
|
#XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
|
|
function1=contenders.lil_devil
|
|
function2=contenders.lil_devil
|
|
|
|
for i in range(nr_of_games):
|
|
helper_functions.play_game( games,function1,function2,nr_of_rounds)
|
|
scores.append(helper_functions.analyze_results(games,len(scores)+1))
|
|
games=[]
|
|
|
|
print("Played "+str(nr_of_games)+ " games with "+str(nr_of_rounds)+" rounds each.")
|
|
print("(A huge disparity of the average and median could hint at special behaviour)")
|
|
helper_functions.analyze_scores(scores) |