Towards the Automatic Identification of Faulty Multi-Agent Based Simulation Runs Using MASTER [chapter]

Chris J. Wright, Phil McMinn, Julio Gallardo
2013 Lecture Notes in Computer Science  
Testing a multi-agent based model is a tedious process that involves generating very many simulation runs, for example as a result of a parameter sweep. In practice, each simulation run must be inspected manually to gain complete confidence that the agent-based model has been implemented correctly and is operating according to expectations. We present MASTER, a tool which aims to semi-automatically detect when a simulation run has deviated from "normal" behaviour. A simulation run is flagged as
more » ... "suspicious" when certain parameters traverse normal bounds determined by the modeller. These bounds are defined in reference to a small series of actual executions of the model deemed to be correct. The operation of MASTER is presented with two case studies, the first with the well-known "flockers" model supplied with the popular MASON agent-based modelling toolkit, and the second a skin tissue model written using another toolkit-FLAME.
doi:10.1007/978-3-642-38859-0_11 fatcat:tyki7xk2qvd6xewf6qgq6soud4