Package tuffy.ground
Provides facilities for grounding MLNs, including KBMC, SQL-based grounding, and MRF partitioning.
See:
Description
Class Summary |
Grounding |
This class handles the grounding process of MLN inference/learning
with SQL queries. |
KBMC |
"Syntactic" Knowledge Base Model Constructor. |
KBMC.AtomCutSet |
AtomCutSet is set of atoms with the same predicate. |
Package tuffy.ground Description
Provides facilities for grounding MLNs, including KBMC, SQL-based grounding, and MRF partitioning.
Grounding will be conducted in three phases:
- Ground relevant atoms: This phase is done by class KBMC. In this phase, a rough sift
of atoms results from a partially grounded predicate tables. In these tables (one
for each predicate), those definitely not useful atoms are not grounded.
- Pick active atoms: This phase is done by class Grounding. In this phase, all
atoms that will not violate any clauses by default are further eliminated
on the basis of the first phase. These eliminated atoms will not play any role
in following inference or learning, therefore grounding them out will make no sense.
- Ground active clauses: This phase is done by class Grounding. In this phase, clauses
are grounded according to picked active atoms in the second phase. Clauses relying on
exactly the same atoms will be merged together to form GClause for efficiency
consideration.