FastLOAS

FastLOAS is a new approach to learning weak constraints in ASP. Its learning task is a restricted version of the Learning from Ordered Answer Sets (LOAS) task used by ILASP, but the methodology is based on FastLAS.

The initial version of FastLOAS has several restrictions:

1. All programs are assumed to be categorical. This means that for any example, when the context of the example is combined with the background knowledge and hypothesis, the resulting program has a single answer set.
2. Only observational predicate learning is supported. This means that no predicate occuring in the modeh declarations should appear in the context of an example, in the background knowledge, or in any modeb or modeo declaration.

As all programs are categorical, ILASP’s notion of brave and cautious orderings coincide, so FastLAS only support brave orderings. These are of the form:

#brave_ordering(order_id@penalty, eg1_id, eg2_id).

order_id is an identifier for the ordering example, penalty is a positive integer denoting the cost of not covering the ordering example, and eg1_id and eg2_id are identifiers for two positive examples (eg1 and eg2, respectively).

The meaning of such an ordering is that the unique answer set of the union of B, H and the context of eg1 should be preferred to the unique answer set of the union of B, H and the context of eg2.

For some example tasks, please see here.

For the source code of FastLOAS, please see here.