Inference

Figure 4: Inference procedures of the compositional modeller

The compositional modelling method presented herein employs a four step inference procedure:

1. Model space construction. The model space is an ATMS that efficiently stores all the participants, relations and model design decisions (represented in the form of relevance and model assumptions) that may be part of the final scenario model, as well as the conditions under which each of these participants and relations must or must not be part of the scenario model.
2. aDCSP construction. The model space contains a number of hard constraints on the participants and relations that may be combined. This inference step extracts such restrictions and translates them into an aDCSP.
3. Inclusion of order-of-magnitude preferences. Preferences are associated with relevance and model assumptions in the scenario space as they reflect the relative appropriateness of these assumptions, resulting in an aDPCSP.
4. Scenario model selection. This inference step solves the aDPCSP. The resulting solutions correspond to scenario models that are consistent according to the domain knowledge and optimise the overall preference with respect to the order-of-magnitude preference calculus.

These four steps correspond to the four squares of the compositional model repository in Figure 4

In this section, each of these inference steps is discussed in detail and illustrated by means of simple examples. The next section contains a more detailed example and shows how this procedure can be applied to a non-trivial ecological modelling domain.