Example 3: Sequential Learning

Example 3 presents a skeleton for sequential learning. Sequential learning, or adaptation, is an update process applied to the conditional probability tables. After a network has been built, sequential learning can be applied during operation in order to maintain the correspondence between the model (conditional probability tables) and the real-world domain.

After the network is loaded in HUGIN, the learning parameters are specified. Then follows the build-up and entering of cases, and finally, the tables are updated and node marginals are printed.

#if X64
using size_t = System.UInt64;
using h_index_t = System.Int64;
#else
using size_t = System.UInt32;
using h_index_t = System.Int32;
#endif
#if H_DOUBLE
using h_number_t = System.Double;
#else
using h_number_t = System.Single;
#endif

using System;
using HAPI;

namespace Example
{
    public class Adapt
    {
        public Adapt(String fileName)
        {
            String netFileName = fileName + ".net";
            Domain d = new Domain(netFileName, new DefaultClassParseListener());
            String logFileName = fileName + ".log";
            d.OpenLogFile(logFileName);
            d.Compile();

            SpecifyLearningParameters(d);
            PrintLearningParameters(d);

            EnterCase(d);

            PrintCase(d);

            d.Propagate(Domain.Equilibrium.H_EQUILIBRIUM_SUM, Domain.EvidenceMode.H_EVIDENCE_MODE_NORMAL);

            d.Adapt();

            d.Initialize();

            PrintNodeMarginals(d);

            d.SaveAsNet("q.net");
        }

        private void SpecifyLearningParameters(Domain d)
        {
            NodeList nl = d.GetNodes();
            h_number_t[] data;

            foreach (DiscreteChanceNode node in nl)
            {
                Table table = node.GetExperienceTable();
                data = new h_number_t[table.GetSize()];
                for (size_t i = 0; i < table.GetSize(); i++)
                    data[i] = 1;
                    table.SetData(data);
            }

            foreach (DiscreteChanceNode node in nl)
            {
                Table table = node.GetFadingTable();
                data = new h_number_t[table.GetSize()];
                for (size_t i = 0; i < table.GetSize(); i++)
                    data[i] = 1;
                table.SetData(data);
            }
        }

        private void PrintLearningParameters(Domain d)
        {
            NodeList nl = d.GetNodes();

            foreach (DiscreteChanceNode dcNode in nl)
            {
                Console.WriteLine(dcNode.GetLabel() + " (" + dcNode.GetName() + "): ");
                Console.Write("   ");
                if (dcNode.HasExperienceTable())
                {
                    Table table = dcNode.GetExperienceTable();
                    h_number_t[] data = table.GetData();
                    size_t tblSize = table.GetSize();

                    for (size_t i = 0; i < tblSize; i++)
                        Console.Write(data[i] + " ");
                    Console.WriteLine();
                }
                else
                    Console.WriteLine("No experience table");

                Console.Write("   ");
                if (dcNode.HasFadingTable())
                {
                    Table table = dcNode.GetFadingTable();
                    h_number_t[] data = table.GetData();
                    size_t tblSize = table.GetSize();

                    for (size_t i = 0; i < tblSize; i++)
                        Console.Write(data[i] + " ");
                    Console.WriteLine();
                }
                else
                    Console.WriteLine("No fading table");
            }
        }

        private void EnterCase(Domain d)
        {
            NodeList nl = d.GetNodes();

            foreach (DiscreteChanceNode dcNode in nl)
            {
                dcNode.SelectState(0);
            }
            DiscreteChanceNode node = (DiscreteChanceNode)nl[1];
            node.RetractFindings();
        }

        private void PrintCase(Domain d)
        {
            NodeList nl = d.GetNodes();

            foreach (DiscreteChanceNode dcNode in nl)
            {
                Console.Write(" (" + dcNode.GetName() + ",");
                if (dcNode.EvidenceIsEntered())
                    Console.Write(" evidence is entered) ");
                else
                    Console.Write(" evidence is not entered) ");
            }
            Console.WriteLine();
        }

        private void PrintNodeMarginals(Domain d)
        {
            NodeList nl = d.GetNodes();

            foreach (DiscreteChanceNode dcNode in nl)
            {
                size_t nStates = dcNode.GetNumberOfStates();

                Console.WriteLine(dcNode.GetLabel() + " (" + dcNode.GetName() + ")");
                for (size_t i = 0; i < nStates; i++)
                    Console.WriteLine(" - " + dcNode.GetStateLabel(i) + ": " + dcNode.GetBelief(i));
            }
        }

        public static void Main(String[] args)
        {
            try
            {
                new Adapt(args[0]);
            }
            catch (ExceptionHugin eh)
            {
                Console.WriteLine(eh);
            }
        }
    }
}