C++ Domain::addAtoms方法代码示例

boost::unordered_set<Model> MCSat::sampleSat(const Model& initialModel, const Domain& d, boost::mt19937& rng) {
    boost::unordered_set<Model> models;
    models.insert(initialModel); // always include the initial model

    // transform domain into a SAT problem
    Domain dSat;
    for (Domain::fact_const_iterator it = d.facts_begin(); it != d.facts_end(); it++) {
        dSat.addFact(*it);
    }
    for (Domain::formula_const_iterator it = d.formulas_begin(); it != d.formulas_end(); it++) {
        ELSentence s = *it;
        s.setHasInfWeight(true);
        dSat.addFormula(s);
    }
    dSat.addAtoms(d.atoms_begin(), d.atoms_end());
    // perform UP (if possible)
    Domain reduced;
    if (useUnitPropagation_) {
        try {
            reduced = MCSat::applyUP(dSat);
        } catch (contradiction& c) {
            return models;  // can't continue, just return our models which has only one item.
        }
    } else {
        reduced = dSat;
    }

    // rewrite infinite weighted formulas so they have singular weight
    reduced = reduced.replaceInfForms();

    // do some random restarts and hope for different models
    MWSSolver walksatSolver(walksatIterations_, walksatRandomMoveProb_, &reduced);
    for (unsigned int i = 1; i <= walksatNumRandomRestarts_; i++) {
        Model iterInitModel = reduced.randomModel(rng);   // TODO: better way to make random models
        if (reduced.formulas_size() == 0) {
            // just add the random model and continue
            models.insert(reduced.randomModel(rng));
            continue;
        }

        //std::cout << "--\nFormulas for maxwalksat: ";
        //std::copy(reduced.formulas_begin(), reduced.formulas_end(), std::ostream_iterator<ELSentence>(std::cout, ", "));
        //std::cout << std::endl;

        Model iterModel = walksatSolver.run(rng, iterInitModel);
        if (reduced.isFullySatisfied(iterModel)) models.insert(iterModel);
    }
    return models;
}

void MCSat::run(boost::mt19937& rng) { // TODO: setup using random initial models
    if (d_ == 0) {
        throw std::logic_error("MCSat::run() - Domain not set");
    }
    if (sampleStrategy_ == 0) {
        throw std::logic_error("MCSat::run() - SampleStrategy not set");
    }
    samples_.clear();
    samples_.reserve(numSamples_);

    //std::cout << "initial domain: ";
    //d_->printDebugDescription(std::cout);

    // first, run unit propagation on our domain to get a new reduced one.
    Domain reduced;
    if (useUnitPropagation_) {
        reduced = MCSat::applyUP(*d_);
    } else {
        reduced = *d_;
    }
    //std::cout << "reduced domain: ";
    //reduced.printDebugDescription(std::cout);


    Model prevModel = (useRandomInitialModels_ ? reduced.randomModel(rng) : reduced.defaultModel());

    // do a starting run on the whole problem as our initial sample
    //boost::unordered_set<Model> initModels = sampleSat(prevModel, reduced);
    //prevModel = *initModels.begin();
    Domain prevDomain = reduced;

    if (burnInIterations_ == 0) samples_.push_back(prevModel);
    unsigned int totalIterations = numSamples_+burnInIterations_;


    for (unsigned int iteration = 1; iteration < totalIterations; iteration++) {
        std::vector<ELSentence> newSentences;

        if ( totalIterations < 20 ||
                iteration % (totalIterations / 20) == 0) {
            std::cout << (((double)iteration) / ((double) totalIterations))*100 << "% done." << std::endl;
        }

        sampleStrategy_->sampleSentences(prevModel, reduced, rng, newSentences);
//        if (iteration == 1) {
//            std::cout << "initial sampled sentences: ";
//            std::copy(newSentences.begin(), newSentences.end(), std::ostream_iterator<ELSentence>(std::cout, "\n"));
//            std::cout << "initial model:";
//            std::cout << prevModel;
//            for (Domain::formula_const_iterator it = prevDomain.formulas_begin();
//                    it != prevDomain.formulas_end();
//                    it++) {
//                std::cout << "formula " << *it << " is satisfied at: " << it->dSatisfied(prevModel, prevDomain) << std::endl;
//            }
//            std::cin.get();
//        }

        // make a new domain using new Sentences
        Domain curDomain;
        curDomain.setMaxInterval(prevDomain.maxInterval());
        for (Domain::fact_const_iterator it = prevDomain.facts_begin(); it != prevDomain.facts_end(); it++) {
            curDomain.addFact(*it);
        }
        for (std::vector<ELSentence>::const_iterator it = newSentences.begin(); it != newSentences.end(); it++) {
            curDomain.addFormula(*it);
        }
        curDomain.addAtoms(prevDomain.atoms_begin(), prevDomain.atoms_end());
//
//        if (iteration == burnInIterations_ + numSamples_/2) {
//            std::cout << "ITERATION: " << iteration << std::endl;
//            std::cout << "curDomain";
//            curDomain.printDebugDescription(std::cout);
//            std::cout << "sampled sentences: ";
//            std::copy(newSentences.begin(), newSentences.end(), std::ostream_iterator<ELSentence>(std::cout, "\n"));
//        }


        boost::unordered_set<Model> curModels = sampleSat(prevModel, curDomain, rng);
        assert(!curModels.empty());
        // choose a random model
        boost::uniform_int<std::size_t> pickModel(0, curModels.size()-1);
        boost::unordered_set<Model>::size_type index = pickModel(rng);
        boost::unordered_set<Model>::const_iterator it = curModels.begin();
        while (index > 0) {
            it++;
            index--;
        }
        // add the model
        if (iteration >= burnInIterations_) samples_.push_back(*it);
        prevModel = *it;
        prevDomain = curDomain;
    }
}