Prisoner's Dilemma - Python

(2018) Attempts to solve the Prisoner's Dilemma problem using genetic algorithms.

Table of Contents

1. About The Project
2. Usage
3. Training
4. License
5. Contact

About The Project

This program was my first attempt at implementing the core concepts of genetic algorithms (fitness -> mutation/crossover -> pruning -> ↵). I would improve my understanding of these concepts when approaching a harder genetic algorithm problem: GitHub: Blackjack-Genetic (teaching agents how to play well at Blackjack using genetic algorithms).

Usage

Prerequisites

• Python 3

Auto generated by Argparse via commit 3871564.

usage: PrisonersDilemma.py [-h] [-p POPULATION] [-g GENERATIONS]
                           [-m MUTATION_RATE] [-t TESTS] [-s SEED]

optional arguments:
  -h, --help        show this help message and exit
  -p POPULATION     Specifies the population size for the group of prisoners.
                    (default: 1024)
  -g GENERATIONS    Specifies the number of generations which should be
                    simulated. (default: 100)
  -m MUTATION_RATE  Specifies the rate in which mutations occur in each
                    genome. (default: 0.15)
  -t TESTS          Specifies the number of tests the prisoners are subjected
                    to per generation. (default: 10)
  -s SEED           Specifies the random number generator seed to be used.
                    (default: 1071815745)

Training

Example Training Parameters

python PrisonersDilemma.py -g 50 -m 0.05 -t 20 -s 23915871

• Population: default (number of prisoners)
• Generations: 50 (number of iterations)
• Mutation Rate: 0.05 (chance [0.0, 1.0] for a gene to be mutated)
• Tests: 20 (number of trials to subject prisoners to the Prisoner's Dilemma)
• Seed: 23915871

---

~~~~~~ Generation #0 ~~~~~~
Average Defection (%)29.83
Average Prisoner Age 0.000
~~~~~~ Generation #1 ~~~~~~
Average Defection (%)30.87
Average Prisoner Age 0.500
~~~~~~ Generation #2 ~~~~~~
Average Defection (%)33.41
Average Prisoner Age 0.753
~~~~~~ Generation #3 ~~~~~~
Average Defection (%)37.82
Average Prisoner Age 0.864
~~~~~~ Generation #4 ~~~~~~
Average Defection (%)44.04
Average Prisoner Age 0.922

Generations #5 -> #45 Omitted (Full Output: Git Gist)

~~~~~~ Generation #45 ~~~~~~
Average Defection (%)96.38
Average Prisoner Age 1.183
~~~~~~ Generation #46 ~~~~~~
Average Defection (%)96.55
Average Prisoner Age 1.144
~~~~~~ Generation #47 ~~~~~~
Average Defection (%)96.75
Average Prisoner Age 1.143
~~~~~~ Generation #48 ~~~~~~
Average Defection (%)95.98
Average Prisoner Age 1.101
~~~~~~ Generation #49 ~~~~~~
Average Defection (%)95.69
Average Prisoner Age 1.168

---

Example Training Takeaways

• Defection rate approaches 100%, however poor mutations stunt further overall-progress.
• Low average age is normal, due to half of the population having an age of 0 (new-born). However it appears as if ancestors are being culled before reaching high ages of maturity. Perhaps with far more generations, the average age would rise.
• Raising the number of Prisoner's Dilemma tests per generation seems to dramatically speed up the learning process. Fewer better-performing prisoners are culled by mistake, and thus the population develops faster.
• By generation 15, over 90% of the population choses to defect. Over the next 35 generations, this is only refined by ~5%.

License

Distributed under the GPL-3.0 License. See LICENSE for more information.

Contact

Kevin Tyrrell - [email protected]

Project Link: https://github.com/KevinTyrrell/PrisonersDilemma-Py