Trust in exchange

This repository contains the sources for running a multi-agent system that models trust in market exchanges.

Running the model

The model can be run with:
python run.py [-h] [-a {MSAgent,WHAgent,RLAgent,GossipAgent,RLGossipAgent}] [-m {[0.0,1.0]}] [-N {[0,10000]}] [-n {[0,10000]}] [-l {[0.0,1.0]}] [-sl {[0.0,1.0]}] [-df {[0.0,1.0]}] [-r {True,False}] [-ms {[0,10000]}] [-t1 {[0,1000000]}] [-t2 {[1,1000000]}] [--save-filename SAVE_FILENAME]

• -h, --help - Show the help message and exit
• -a, --agent-class - {MSAgent, WHAgent, RLAgent, GossipAgent, RLGossipAgent} - Which type of agent to use.
• -m, --mobility-rate - [0.0,1.0] - The probability of an agent moving to a new neighbourhood.
• -N, --number-of-agents - [0,10000] - The total number of agents in the model.
• -n, --neighbourhood-size - [0,10000] - The initial number of agents in each neighbourhood.

RLAgent, RLGossipAgent only:

• -l, --learning-rate - [0.0,1.0] - (RLAgent, RLGossipAgent only) The discount factor with which the probabilities are updated.
• -sl, --social-learning-rate - [0.0,1.0] - (RLAgent, RLGossipAgent only) The probability of copying a propensity.
• -df, --discount-factor - [0.0,1.0] - (RLAgent, RLGossipAgent only) The discount factor of the cumulative reward.
• -r, --relative-reward - {True, False} - (RLAgent, RLGossipAgent only) Whether to normalize rewards to a mean of zero.

GossipAgent, RLGossipAgent only:

• -ms, --memory-size - [0,10000] - (GossipAgent, RLGossipAgent only) The number of memories an agent can store.

Running model parameters:

• -t1, --T_onset - [0,1000000] - The number of time steps to run before recording data.
• -t2, --T_record - [1,1000000] - The number of time steps to run for recording the data.
• --save-filename - SAVE_FILENAME - Saves to /m_SAVE-FILENAME and /a_SAVE-FILENAME

Repository contents description

• The starting point for running the code is the file run.py. runMultipleExperiments.py contains the code for running several experiments.
• The model implementation can be found in the trust folder. The utils folder contains some utilties for use by the model and running scripts.
• The data from running the model with run.py will be stored in the data folder.
• Scripts for plotting and some resulting plots is to be found in the plotting folder.