zvyagintsev / light-perceptron

LightPerceptron is a simple Java library that allows you to create and train multilayer perceptrons.

LightPerceptron

The perceptron is an algorithm for supervised learning of binary classifiers. Perceptron was one of the first models of neural networks.

LightPerceptron is a simple Java library that allows you to create and train multilayer perceptrons.

Learning algorithm: backpropagation

Activation functions:

• SIGMOID: f(x) = 1 / (1 - exp(-x))

LightPerceptron also allows you to import / export trained neural networks into json files.

build from sources:

cd light-perceptron
mvn install

How to use:

Creating network with PerceptronBuilder:

        Perceptron perceptron = new PerceptronBuilder().
                addLayer(2, 2).
                addLayer(2, 1).
                build();

This code created perceptron with with 2 layers (hidden with 2 neurons and output layer with 1 neuron).

Sample 1: Animal classification task

number of wings	number of paws	has tail	insect	bird	snake	mammal	spider
2	6	0	1	0	0	0	0
2	2	1	0	1	0	0	0
0	0	1	0	0	1	0	0
0	4	0	0	0	0	1	0
0	8	0	0	0	0	0	1

final double[][] DATA_ANIMALS = {
        {2, 6, 0, 1, 0, 0, 0, 0}, // insect
        {2, 2, 1, 0, 1, 0, 0, 0}, // bird
        {0, 0, 1, 0, 0, 1, 0, 0}, // snake
        {0, 4, 0, 0, 0, 0, 1, 0}, // mammal
        {0, 8, 0, 0, 0, 0, 0, 1}  // spider

};
PerceptronBuilder pb = new PerceptronBuilder();
Perceptron p = pb.
        addLayer(3, 10).
        addLayer(10, 10).
        addLayer(10, 5).
        build();

p.training(DATA_ANIMALS, 3, 5, 10000);
double[] result = p.classification(new double[] {0, 4, 1});

Sample 2: Function approximation

        int maxIteration = 1000;
        PerceptronBuilder pb = new PerceptronBuilder();
        Perceptron p = pb.
                addLayer(2, 2).
                addLayer(2, 1).
                build();

        Random r = new Random();
        double a, b;

        for (int i = 0; i < maxIteration; i++) {
            // generate a, b, 0 <= a <= 1, 0 <= a <=
            a = r.nextDouble();
            b = r.nextDouble();
            p.training(
                    new double[]{a, b}, new double[]{Math.sin(a) * Math.cos(b)});
        }
        a = r.nextDouble();
        b = r.nextDouble();
        double[] calcResult = p.calculate(new double[] {a, b});

Import / export to json format

sample code with import and export of perceptron state:

        final double[][] DATA_ANIMALS = {
            {2, 6, 0, 1, 0, 0, 0, 0}, // insect
            {2, 2, 1, 0, 1, 0, 0, 0}, // bird
            {0, 0, 1, 0, 0, 1, 0, 0}, // snake
            {0, 4, 0, 0, 0, 0, 1, 0}, // mammal
            {0, 8, 0, 0, 0, 0, 0, 1}  // spider
        };

        PerceptronBuilder pb = new PerceptronBuilder();
        Perceptron p = pb.
                addLayer(3, 10).
                addLayer(10, 10).
                addLayer(10, 5).
                build();


        // training neural net
        p.training(DATA_ANIMALS, 3, 5, 1000);
        
        double[] result = p.classification(new double[] {2, 6, 0});
        String str = p.exportToJson();

        Perceptron p2 = pb.loadFromJson(str);
        double[] result2 = p2.classification(new double[] {2, 6, 0});

Json with a perceptron state:

{
  "layers": [
    {
      "nInput": 3,
      "nOutput": 5,
      "neurons": [
        {
          "weights": [
            -1.7874804103060071,
            1.3755867550799334,
            -0.07769751578997851
          ],
          "threshold": -1.5545014269000352
        },
        {
          "weights": [
            -1.2657621818702498,
            -0.008026094042472627,
            1.817357363951599
          ],
          "threshold": 0.3079024314435577
        },
        {
          "weights": [
            2.3869309691959013,
            -1.6920546412863395,
            1.005298869863273
          ],
          "threshold": 1.5551406483323806
        },
        {
          "weights": [
            2.720545987044474,
            -0.05007972364363105,
            -2.1874063960889223
          ],
          "threshold": -1.1972588963891837
        },
        {
          "weights": [
            1.2029037228992203,
            1.76800111428798,
            -1.4732765710439788
          ],
          "threshold": -1.4070653175289247
        }
      ]
    },
    {
      "nInput": 5,
      "nOutput": 4,
      "neurons": [
        {
          "weights": [
            1.594427083203863,
            -1.8268179641073674,
            -3.6207943250994283,
            3.4972727771716685,
            -0.367092267344614
          ],
          "threshold": -2.1331673348798663
        },
        {
          "weights": [
            -2.9727129035197613,
            -2.1181851676309216,
            2.463672856513635,
            1.5202010079774575,
            1.7132122484985965
          ],
          "threshold": -2.71614972112006
        },
        {
          "weights": [
            -1.3486129625435426,
            0.6213533111488375,
            2.230717431733854,
            -2.0890523941773504,
            -3.491911158294164
          ],
          "threshold": 0.5714130803747863
        },
        {
          "weights": [
            1.7636905875821767,
            0.9143079497423531,
            -2.805789583056562,
            -4.097382327580715,
            0.8746795299254279
          ],
          "threshold": -1.001793049363282
        }
      ]
    }
  ]
}