beliefpropagation.h

gibbs

#ifndef BELIEFPROPAGATION_H
#define BELIEFPROPAGATION_H

#include "InferenceMethod.h"
#include "MarkovNet.h"
#include "BayesNet.h"

class BeliefPropagation : public InferenceMethod
{
private:
    const BayesNet& bnet;
    MarkovNet model;
    double convergenceThreshold;
    double damping;

public:
    // Simple constructor
    BeliefPropagation(const BayesNet& bn, double ratio) 
        : bnet(bn), model(bn), convergenceThreshold(ratio), damping(0.0) 
     { /* NOP */ }

    virtual ~BeliefPropagation() { /* NOP */ }

    virtual void runMarginalInference(const VarSet& evidence)
    {
        // Reset network
        model.resetAllNodes();
        model.resetAllMessages();

        // Fix evidence variables
        for (int i = 0; i < evidence.getNumVars(); i++) {
            if (evidence.isTested(i)) {
                // All network variables must be discrete.
                model.fixNodeValue(i, (int)evidence[i]);
            }
        }

        // Run belief propagation to convergence
        model.runBP(convergenceThreshold, damping);
    }

    virtual double singleConditionalLogProb( const list<int>& queryVars,
        const VarSet& evidence, const VarSet& answer) const { return 0; }

#if 0
    virtual Distribution getMarginal(int i) const {
        return model.getMarginal(i);
    }
#endif

    virtual double getMarginalProb(int i, double value) const {
        Distribution m = model.getMarginal(i);
        return m[(int)value];
    }

    virtual int getRange(int i) const { return bnet.getRange(i); }

    virtual void runJointInference(const list<int>& queryVars,
            const VarSet& evidence) { 
        cout << "ERROR: joint inference not supported "
            << "using BeliefPropagation!\n";  
    }

    virtual double getJointProb(const VarConfig& query) const {
        return 0;
    }
};

#endif // ndef BELIEFPROPAGATION_H