Pandora: A versatile agent-based modelling platform for social simulation

The evolving field of social simulation is diversifying the degree of complexity of published works, from simple models to large scale simulations with millions of agents. In this context, available platforms are divided between the ones favoring easy to use interfaces and the development of prototyping, and others created for simulating large scenarios with high computing costs. The later group is truly diverse, based on the way executions are accelerate and the wide range of technologies that they can support. The cost of this increase in performance is a steep learning curve, as the users of these platforms need to have advanced programming skills in order to deal with code complexity. Pandora is a novel open-source framework designed to fill the gap between these approaches. A twin interface in Python and C++ offers the same interface to users developing prototypes as well as projects with High-Performance Computing requirements. The need for parallel programming knowledge is also skipped through the automated generation, during compilation time, of needed code for shared and unshared memory distribution using OpenMP and MPI. A set of different helpers (unit testing, georeferencing support) and analytical tools complement the basic framework in order to facilitate the tasks of development, testing and analyzing any type of Agent-Based Model. Pandora's flexibility is exemplified through different projects that have introduced GPU acceleration, georeferenced data and cloud computing to the framework.