Package tuffy.infer

Provides MLN inference algorithms.


Class Summary
DataMover Methods for moving data around between the RDBMS and memory, as well as writing inference results to files.
InferBucket A bucket of inference tasks that can run in prallel.
InferBucket.CompWorker A worker thread that runs inference on one component at a time.
InferComponent Performing inference on one MRF component.
InferPartitioned Scheduler of partition-aware inference.
MRF In-memory data structure representing an MRF.
PGInfer Deprecated.  

Enum Summary

Package tuffy.infer Description

Provides MLN inference algorithms. Currently supports MAP inference and Marginal inference.

By default this package deals with inference in memory. It first loads from database about information of MLN groundings. Then, it forms a MRF by taking GAtom and GClause as elements. Once this MRF is formed, this package will not interact with DBMS.

Marginal and MAP inference on this in-memory structure follow classic scheme. There are indices from GClause to GAtom, vice versa. Therefore, the change of GAtom's truth assignment will influence GClause's truth value. Together with the sign of GClause's weight, whether a GClause is violated in current truth assignment setting can be tracked. The WalkSAT or MCSAT subroutines are then invoked.