initinference.cpp

gibbs

#include "InferenceMethod.h"
#include "GibbsSampler.h"
#include "DummyInference.h"
#include "BeliefPropagation.h"
#include "NaiveBayesInference.h"

#include <iostream>
#include <fstream>
#include "VarSchema.h"

InferenceMethod* initDummyInference(int argc, char* argv[], int minArg)
{
    ifstream in(argv[minArg]);

    if (!in) {
        cout << "ERROR: could not open schema \"" << argv[minArg] << "\"\n";
        exit(-1);
    }

    VarSchema schema;
    in >> schema;
    return new DummyInference(schema);
}

InferenceMethod* initGibbsSampler(int argc, char* argv[], int minArg)
{
    FILE* inFile = fopen(argv[minArg], "r");

    if (!inFile) {
        cout << "ERROR: could not open model file \"" << argv[minArg] << "\"\n";
        exit(-1);
    }

    BayesNet* model = new BayesNet();
    loadModel(inFile, *model);

    // Start out with reasonable defaults
    int numChains = 10;
    int burnInIters = 100;
    int minIters = 1000;
    double ratio = 1.1;
    bool fixedIters = false;

    // Adjust them as indicated by arguments
    for (int i = minArg+1; i < argc; i++) {
        if (argv[i][0] != '-') {
            cout << "Skipping unknown parameter: " << argv[i] << endl;
            continue;
        }

        switch (argv[i][1]) {
            case 'c': numChains = atoi(argv[++i]); break;
            case 'r': ratio = atof(argv[++i]); break;
            case 'b': burnInIters = atoi(argv[++i]); break;
            case 'm': minIters = atoi(argv[++i]); break;
            case 'f': fixedIters = true; break;
            default: cout << "Skipping unknown parameter: " << argv[i] << endl;
        }
    }
    
    return new GibbsSampler(*model, numChains, ratio, burnInIters, minIters,
            fixedIters);
}

InferenceMethod* initBeliefPropagation(int argc, char* argv[], int minArg)
{
    FILE* inFile = fopen(argv[minArg], "r");

    if (!inFile) {
        cout << "ERROR: could not open model file \"" << argv[minArg] << "\"\n";
        exit(1);
    }

    BayesNet model;
    loadModel(inFile, model);

    // Start out with reasonable defaults
    double ratio = 1.1;

    // Adjust them as indicated by arguments
    for (int i = minArg+1; i < argc; i++) {
        if (argv[i][0] != '-') {
            cout << "Skipping unknown parameter: " << argv[i] << endl;
            continue;
        }

        switch (argv[i][1]) {
            case 'r': ratio = atof(argv[++i]); break;
            default: cout << "Skipping unknown parameter: " << argv[i] << endl;
        }
    }
    
    return new BeliefPropagation(model, ratio);
}

#include "../src/AbsVarAbstractionSet.h"
InferenceMethod* initNaiveBayesInference(int argc, char* argv[], int minArg)
{
    ifstream in(argv[minArg]);

    if (!in) {
        cout << "ERROR: could not open model file \"" << argv[minArg] << "\"\n";
        exit(-1);
    }

    int modelType;
    in >> modelType;
    if (!in || modelType != 3) {
        cout << "Invalid model file \"" << argv[minArg] << "\"\n";
        exit(-1);
    }
    AbstractionSet* model = new AbsVarAbstractionSet;
    in >> *model;
    return new NaiveBayesInference(model);
}

InferenceMethod* initInferenceMethod(int argc, char* argv[], int minArg)
{
    InferenceMethod* method = NULL;

    switch(argv[minArg][1]) {
        // Baseline "dummy" algorithm
        case 'd': method = initDummyInference(argc, argv, minArg+1);
                  break;

        // Gibbs sampling
        case 'g': method = initGibbsSampler(argc, argv, minArg+1);
                  break;

        // Belief propagation
        case 'b': method = initBeliefPropagation(argc, argv, minArg+1);
                  break;

        // Naive Bayes
        case 'n': method = initNaiveBayesInference(argc, argv, minArg+1);
                  break;

        default:  cout << "Unknown inference method \"" << argv[minArg] << "\"\n";
                  exit(-1);
                  break;
    }

    return method;
}