willis-richard / reward_transfer_matrix Goto Github PK

This repository accompanies the paper "Resolving social dilemmas with minimal reward transfer", arXiv, containing the code required to reproduce its results.

Install the dependencies using conda

conda create -f environment.yml

Activate the environment and add the repo to the PYTHONPATH

conda activate rtm
export PYTHONPATH="$(pwd)/src"

src.rtm.algorithms contains Algorithm 1 (called find_T_star), which returns a minimal reward transfer matrix for a normal-form social dilemma. It also has a simpler algorithm to find the symmetrical self-interest level of a game, and algorithm to maximise the off-diagonal entropy of a reward transfer matrix.

In order to produce the results from the paper, call

python3 src/rtm/solver_lp.py --game <GAME_NAME>

To see the options, use

python3 src/rtm/solver_lp.py --help