tuffy.ground.partition
Class Partitioning

java.lang.Object
  tuffy.ground.partition.Partitioning

public class Partitioning
extends java.lang.Object

Utilities for partitioning the MRF generated by the grounding process.

Field Summary

private RDB	db
private MarkovLogicNetwork	mln

Constructor Summary

Partitioning(Grounding g)
Construct a partitioning worker based on the grounding result.

Method Summary

PartitionScheme	partitionAtoms(double ramBudgetPerPartition) Agglomeratively cluster the atoms in the MRF into partitions while heuristically minimizing the (weighted) cut size.
void	partitionClauses(PartitionScheme pmap) Given a partitioning scheme, partition the data accordingly.
PartitionScheme	partitionMRF(double maxPartitionSize)
private double	perAtomWeight()
private double	perClauseSharedWeight(int nlits)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

mln

private MarkovLogicNetwork mln

db

private RDB db

Constructor Detail

Partitioning

public Partitioning(Grounding g)

Construct a partitioning worker based on the grounding result.

Parameters:

g - the grounding worker

Method Detail

perAtomWeight

private double perAtomWeight()

perClauseSharedWeight

private double perClauseSharedWeight(int nlits)

partitionAtoms

public PartitionScheme partitionAtoms(double ramBudgetPerPartition)

Agglomeratively cluster the atoms in the MRF into partitions while heuristically minimizing the (weighted) cut size. The heuristic is to scan through all clauses in the descending-abs(weight) order.

Parameters:

ramBudgetPerPartition - size bound of a partition, roughly proportional to the number of atoms and clauses in each partition

Returns:

a partitioning scheme

partitionMRF

public PartitionScheme partitionMRF(double maxPartitionSize)

partitionClauses

public void partitionClauses(PartitionScheme pmap)

Given a partitioning scheme, partition the data accordingly.