Creating a Kohonen Self-Organizing Network

The example shows how one should classify countries according to the assigned set of characteristics, using a Kohonen network. Data about countries is implemented by the Country class.

To execution the example, add links to the NN and Collections system assemblies.

Class Country: Object
    Public Name: String;
    Public GDP: Double;
    Public Population: Double;
    Public PopulationGrowthRate: Double;
    Public AreaLand: Double;
    Public AreaWater: Double;
    Public IrrigatedLand: Double;

    Public Constructor Create(_Name: String; _GDP: Double; _Population: Double; _PopulationGrowthRate: Double; _AreaLand: Double; _AreaWater: Double; _IrrigatedLand: Double);
    Begin
        Name := _Name;
        GDP := _GDP;
        Population := _Population;
        PopulationGrowthRate := _PopulationGrowthRate;
        AreaLand := _AreaLand;
        AreaWater := _AreaWater;
        IrrigatedLand := _IrrigatedLand;
    End Constructor Create;

    Public Constructor CreateEmpty;
    Begin
    End Constructor CreateEmpty;
End Class Country;

Function Max(D1, D2: Double): Double;
Begin
    If D1 > D2 Then
        Return D1;
    End If;
    Return D2;
End Function Max;

// Method for initialization of the network
Sub NetworkInitialize(Network: NeuralNetwork);
Var
    numberOfClusters: Integer =3;
    numberOfNeurons: array[0..0] Of Integer;
Begin
    numberOfNeurons[0] := numberOfClusters;
    Network.CreateNetworkEx(1, 6, numberOfNeurons, 0.5, 3);
    Network.InitSynapsesConvex;
End Sub NetworkInitialize;

Sub Main;
Var
    network: NeuralNetwork;
    countries : ArrayList;
    normalizedCountries: ArrayList;
    i, epoch: Integer;
    normalizedCountry, currCountry: Country;
    inputs : array[0..5] Of Double;
    alfa : Double;
    learnRate : Double;
    learnRadius: Integer;
    sampleClass: Integer;
Begin

// Assign initial values
    countries := New ArrayList.Create;
    countries.Add(New Country.Create("Algeria", 3.9, 31193917, 1.74, 2381740, 0, 5550));
    countries.Add(New Country.Create("Argentina", -3, 36955182, 1.16, 2736690, 30200, 17000));
    countries.Add(New Country.Create("Brazil", 0.8, 172860370, 0.94, 8456510, 55455, 28000));
    countries.Add(New Country.Create("Canada", 3.6, 31281092, 1.02, 9220970, 755170, 7100));
    countries.Add(New Country.Create("China", 7, 1261832482, 0.9, 9326410, 270550, 498720));
    countries.Add(New Country.Create("India", 5.5, 1014003817, 1.58, 2973190, 314400, 480000));
    countries.Add(New Country.Create("Indonesia", 0, 224784210, 1.63, 1826440, 93000, 45970));
    countries.Add(New Country.Create("Iran", 1, 65619636, 0.83, 1636000, 12000, 94000));
    countries.Add(New Country.Create("Kazakhstan", 1.7, 16733227, -0.05, 2669800, 47500, 22000));
    countries.Add(New Country.Create("Libya", 2, 5115450, 2.42, 1759540, 0, 4700));
    countries.Add(New Country.Create("Mexico", 3.7, 100349766, 1.53, 1923040, 49510, 61000));
    countries.Add(New Country.Create("Mongolia", 3.5, 2650952, 1.54, 1565000, 0, 800));
    countries.Add(New Country.Create("Niger", 2, 10075511, 2.75, 1266700, 300, 660));
    countries.Add(New Country.Create("Peru", 2.4, 27012899, 1.75, 1280000, 5220, 12800));
    countries.Add(New Country.Create("Russia", 3.2, 146001176, -0.38, 16995800, 79400, 40000));
    countries.Add(New Country.Create("Saudi Arabia", 1.6, 22023506, 3.28, 1960582, 0, 4350));
    countries.Add(New Country.Create("Sudan", 3, 35079814, 2.84, 2376000, 129810, 19460));
    countries.Add(New Country.Create("United States", 4.1, 275562673, 0.91, 9158960, 470131, 207000));
    normalizedCountry := New Country.CreateEmpty;
    For i := 0 To countries.Count - 1 Do
        currCountry := countries.Item(i) As Country;
        normalizedCountry.GDP := Max(normalizedCountry.GDP, currCountry.GDP);
        normalizedCountry.Population := Max(normalizedCountry.Population, currCountry.Population);
        normalizedCountry.PopulationGrowthRate := Max(normalizedCountry.PopulationGrowthRate, currCountry.PopulationGrowthRate);
        normalizedCountry.AreaLand := Max(normalizedCountry.AreaLand, currCountry.AreaLand);
        normalizedCountry.AreaWater := Max(normalizedCountry.AreaWater, currCountry.AreaWater);
        normalizedCountry.IrrigatedLand := Max(normalizedCountry.IrrigatedLand, currCountry.IrrigatedLand);
    End For;
    normalizedCountries := New ArrayList.Create;
    For i := 0 To countries.Count - 1 Do
        currCountry := countries.Item(i) As Country;
        normalizedCountries.Add(New Country.Create(currCountry.Name, currCountry.GDP / normalizedCountry.GDP, currCountry.Population / normalizedCountry.Population,
            currCountry.PopulationGrowthRate / normalizedCountry.PopulationGrowthRate, currCountry.AreaLand / normalizedCountry.AreaLand,
            currCountry.AreaWater / normalizedCountry.AreaWater, currCountry.IrrigatedLand / normalizedCountry.IrrigatedLand));
    End For;

// Create a neural network
    network := New NeuralNetwork.Create;
// Initialize network
    NetworkInitialize(network);
// Learn network
    For epoch := 1 To 300 Do
        For i := 0 To normalizedCountries.Count - 1 Do
            currCountry := normalizedCountries.Item(i) As Country;
            inputs[0] := currCountry.GDP;
            inputs[1] := currCountry.Population;
            inputs[2] := currCountry.PopulationGrowthRate;
            inputs[3] := currCountry.AreaLand;
            inputs[4] := currCountry.AreaWater;
            inputs[5] := currCountry.IrrigatedLand;
            alfa := epoch / 100;
            network.SetInputValuesConvex(inputs, alfa);
            learnRate := 0.6 * ((300 - epoch) / 300);
            learnRadius := (3 * ((300 - epoch) / 350)) As Integer;
            network.SetLearnRadius(learnRadius);
            network.SetLearnRate(learnRate);
            network.PropagateSOFM;
            network.LearnSOFM;
        End For;
    End For;

// Test input values
    For i := 0 To normalizedCountries.Count - 1 Do
        currCountry := normalizedCountries.Item(i) As Country;
        inputs[0] := currCountry.GDP;
        inputs[1] := currCountry.Population;
        inputs[2] := currCountry.PopulationGrowthRate;
        inputs[3] := currCountry.AreaLand;
        inputs[4] := currCountry.AreaWater;
        inputs[5] := currCountry.IrrigatedLand;
        alfa := 1;
        network.SetInputValuesConvex(inputs, alfa);
        network.PropagateSOFM;
        sampleClass := network.GetClosestNeuron;
        Debug.WriteLine(CurrCountry.Name + ". Class: " + sampleClass.ToString);
    End For;
// Delete network
    network.DeleteNetwork;
End Sub Main;

After executing the example each country, which data is tested, is related to one of the three classes. Results are displayed in the console window.