/*******************************************************************************
* Copyright (C) 2008-2012 Dominik Jain.
*
* This file is part of ProbCog.
*
* ProbCog is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* ProbCog is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with ProbCog. If not, see <http://www.gnu.org/licenses/>.
******************************************************************************/
package probcog.srl.directed.bln.py;
import java.util.Collection;
import java.util.Vector;
import probcog.bayesnets.core.BeliefNetworkEx;
import probcog.srl.Database;
import probcog.srl.Signature;
import probcog.srl.directed.RelationalBeliefNetwork;
import probcog.srl.directed.bln.AbstractGroundBLN;
import edu.ksu.cis.bnj.ver3.core.BeliefNode;
import edu.ksu.cis.bnj.ver3.core.CPF;
import edu.ksu.cis.bnj.ver3.core.CPT;
import edu.ksu.cis.bnj.ver3.core.Discrete;
import edu.ksu.cis.bnj.ver3.core.values.ValueDouble;
import edu.tum.cs.util.Stopwatch;
public class GroundBLN extends AbstractGroundBLN {
protected BeliefNetworkEx groundBN;
protected Vector<String> hardFormulaNodes;
protected Database db;
public GroundBLN(BayesianLogicNetworkPy bln, String databaseFile) throws Exception {
super(bln, databaseFile);
}
public GroundBLN(BayesianLogicNetworkPy bln, Database db) throws Exception {
super(bln, db);
}
@Override
public void groundFormulaicNodes() throws Exception {
BayesianLogicNetworkPy bln = (BayesianLogicNetworkPy)this.bln;
// ground formulaic nodes
System.out.println(" grounding formulas...");
bln.generateGroundFormulas(this.databaseFile);
GroundFormulaIteration gfIter = bln.iterGroundFormulas();
System.out.printf(" %d formulas instantiated\n", gfIter.getCount());
System.out.println(" instantiating nodes and CPFs...");
Stopwatch sw_structure = new Stopwatch();
Stopwatch sw_cpt = new Stopwatch();
for(GroundFormula gf : gfIter) {
// create a node for the ground formula
sw_structure.start();
String nodeName = "GF" + gf.idxGF;
System.out.println(nodeName + ": " + gf);
Vector<String> GAs = gf.getGroundAtoms();
BeliefNode node = addHardFormulaNode(nodeName, GAs);
// add edges from ground atoms
sw_structure.stop();
// fill CPT according to formula semantics
// TODO try to reuse CPFs generated for previous formulas with same formula index
sw_cpt.start();
fillFormulaCPF(gf, node.getCPF(),GAs);
sw_cpt.stop();
}
System.out.println(" structure time: " + sw_structure.getElapsedTimeSecs() + "s");
System.out.println(" cpf time: " + sw_cpt.getElapsedTimeSecs() + "s");
}
/**
* adds a node corresponding to a hard constraint to the network - along with the necessary edges
* @param nodeName name of the node to add for the constraint
* @param parentGAs collection of names of parent nodes/ground atoms
* @return the node that was added
* @throws Exception
*/
public BeliefNode addHardFormulaNode(String nodeName, Collection<String> parentGAs) throws Exception {
BeliefNode[] domprod = new BeliefNode[1+parentGAs.size()];
BeliefNode node = groundBN.addNode(nodeName);
domprod[0] = node;
hardFormulaNodes.add(node.getName());
int i = 1;
for(String strGA : parentGAs) {
BeliefNode parent = groundBN.getNode(strGA);
if(parent == null) { // if the atom cannot be found, e.g. attr(X,Value), it might be a functional, so remove the last argument and try again, e.g. attr(X) (=Value)
String parentName = strGA.substring(0, strGA.lastIndexOf(",")) + ")";
parent = groundBN.getNode(parentName);
if(parent == null)
throw new Exception("Could not find node for ground atom " + strGA);
}
domprod[i++] = parent;
groundBN.connect(parent, node, false);
}
((CPT)node.getCPF()).buildZero(domprod, false); // ensure correct ordering in CPF
return node;
}
/**
* fills the CPF of a formulaic node
* @param gf the ground formula to evaluate for all possible settings
* @param cpf the CPF of the formulaic node to fill
* @param parents the parents of the formulaic node
* @param parentGAs the ground atom string names of the parents (in case the node names do not match them)
* @throws Exception
*/
protected void fillFormulaCPF(GroundFormula gf, CPF cpf, Vector<String> parentGAs) throws Exception {
BeliefNode[] nodes = cpf.getDomainProduct();
int[] addr = new int[nodes.length];
fillFormulaCPF(gf, cpf, parentGAs, 1, addr);
}
protected void fillFormulaCPF(GroundFormula gf, CPF cpf, Vector<String> parentGAs, int iDomProd, int[] addr) throws Exception {
BeliefNode[] domprod = cpf.getDomainProduct();
// if all parents have been set, determine the truth value of the formula and
// fill the corresponding column of the CPT
State state = ((BayesianLogicNetworkPy)bln).getState();
if(iDomProd == domprod.length) {
// get truth value of formula
double value = gf.isTrue(state) ? 1 : 0;
/*
for(String ga : parentGAs)
System.out.print(ga + " = " + state.get(ga) + ", ");
System.out.println(" -> " + value);
*/
// write to CPF
// - true
addr[0] = 0;
cpf.put(addr, new ValueDouble(value));
// - false
addr[0] = 1;
cpf.put(addr, new ValueDouble(1.0-value));
return;
}
// otherwise get the next ground atom and consider all of its groundings
BeliefNode parent = domprod[iDomProd];
String parentGA = parentGAs.get(iDomProd-1);
Discrete domain = (Discrete)parent.getDomain();
boolean isBoolean = RelationalBeliefNetwork.isBooleanDomain(domain);
// - get the domain index that corresponds to setting the atom to true
int trueIndex = 0;
if(!isBoolean) {
int iStart = parentGA.lastIndexOf(',')+1;
int iEnd = parentGA.lastIndexOf(')');
String outcome = parentGA.substring(iStart, iEnd);
trueIndex = domain.findName(outcome);
if(trueIndex == -1)
throw new Exception("'" + outcome + "' not found in domain of " + parentGA);
}
// - recursively consider all settings
for(int i = 0; i < domain.getOrder(); i++) {
// set address
addr[iDomProd] = i;
// set state for logical reasoner
if(i == trueIndex)
state.set(parentGA, true);
else
state.set(parentGA, false);
// recurse
fillFormulaCPF(gf, cpf, parentGAs, iDomProd+1, addr);
}
}
/**
* gets the ground Bayesian network
* @return
*/
public BeliefNetworkEx getGroundBN() {
return this.groundBN;
}
@Override
protected void onAddGroundAtomNode(BeliefNode instance, String[] params,
Signature sig) {
// TODO Auto-generated method stub
}
@Override
protected void onAddEvidenceVariable(String functionName, String[] params) {
// TODO Auto-generated method stub
}
}