leloykun / sequential-game-theory-simulations

A study on the social behavior of reinforcement learning agents

Simulations of Sequential Games with Multi-agent Reinforcement Learning

Simulations on the collective behavior of reinforcement learning agents

NOTE: This project is not finished yet. See the older version here.

Gettting started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Language

Python 3.x (preferably python 3.5)

Installation

Download or clone the repository using the big green button on the upper-right corner of the page or by using git:

git clone git://github.com/leloykun/socialsims.git

or

git clone https://github.com/leloykun/socialsims.git

Libraries

Install the required libraries with pip:

pip install -r requirements.txt

or

pip3 install -r requirements.txt

Running the simulations

You can run all of the simulations in one go with:

python main.py

The raw data for each simulation can be found in data/raw/[name of simulation]

Testing

Just run

pytest

Turning on the visuals

The visuals are turned off by default to speed up the simulations.

To turn them on, simply uncomment the line with env.show() in the source code of each simulation:

# env.show()
env.show()

This is NOT recommended.

TODO: centralize this option

Running the simulations individually

You can comment out the simulations in src/sims/list.txt to exclude them from being run by portal.py. For example, with the following, only simulation cat-mouse would be run with parameters 100 10 10 as [runs] [trials] [steps].

# cat-mouse [runs] [trials] [steps]
cat-mouse 10 100 10

# cat-mouse-cheese [runs] [timesteps] [interval]
# cat-mouse-cheese 10 10000 100

# simple-migration [runs] [timesteps] [temp_powers]
# simple-migration 10 10000 5

# route-choice [runs] [timesteps] [num_drivers] [[road capacities]]
# route-choice 10 1000 100 10 20 30 40

# migration [runs] [timesteps] [num_mice]
# migration 1 100000 3

TODO: fix this

Data Analysis

TODO

Details

TODO

TODO move this part to 'wiki'

Reinforcement Learning

TODO

Simulations on Specialization

TODO

Simulations on Migration

TODO

Simulations on Cooperation

TODO

Results and Discussion

TODO

TODO use seaborn facet grid to graph data

Developer

• Franz Louis Cesista
• Grade 12 student from Philippine Science High School - Eastern Visayas Campus
• Machine Learning enthusiast

sequential-game-theory-simulations's Issues

cat-mouse-cheese simulations

cat-mouse simulations

mathematical modeling of performance vs. variables

Simulation on Learning