From: Dynamic agent composition for large-scale agent-based models
Shortfalls of existing model building approaches and software systems | MODAM solutions |
---|---|
• Central simulation class is responsible for | • Decentralised factories create the physical properties of an agent (assets) and its behavioural properties (behaviours) separately |
a. Instantiating all agents; | |
b. Defining relationships between instantiated agents; | |
c. Reading the data used to build agents and relationships (if the model is data-driven); | a. Several asset factories create assets and the relationships between them; |
b. Several behaviour factories attach behaviours to assets; | |
d. Complex option handling done as centralised code. | |
c. Each asset factory can be independently parameterised with data providers; | |
d. Each factory handles its own options. | |
• Each agent is a single class. | • Separation of assets and behaviours allows |
a. Variation of behaviour requires many subclasses; | a. Mix-and-match construction of agents at runtime; |
b. Each agent is created independently of others. | b. Gathering certain entities in groups. |
• Non deterministic order of agents’ execution | • Deterministic simulation runs for each random seed (reproducibility of results) |