Applying Dependency Injection to Agent-Based Modeling: the JABM Toolkit

In many areas of science, agent-based models have become increasingly important. These models are often sufficiently complex that deriving closed-form solutions for quantitative aspects of their macroscopic behaviour is often impractical if not impossible. Thus these models are often analysed using Monte-Carlo simulation. Object-oriented programming languages are naturally suited to implementing such models. However, Monte-Carlo methods impose some subtle requirements: we must run the same program very many times with randomly-drawn values of variables in the simulation being drawn from probability distributions, taking great care to ensure that each run is not contaminated with state from previous runs. Typically these problems are tackled by model-specific application code residing in the objects representing individuals in the model. However, this approach is not declarative and leads to crosscontamination between code representing experiments or treatments and code representing the model itself. In this paper we give an overview of agent-based modelling with an emphasis on its application to multi-agent systems research, and show how a design pattern called “dependency injection” can be used to implement highly configurable simulation models which are able to incorporate various assumptions about agents’ rationality and learning.