encog/tests/neural/training/propagation/resilient.specs.js

Encog is a NodeJs ES6 framework based on the Encog Machine Learning Framework by Jeff Heaton, plus some the of basic data manipulation helpers.

describe('ResilientPropagation', function () {
    const Encog = require(PATHS.BASE);
    Encog.Log.options.logLevel = 'silent';

    const ResilientPropagation = Encog.Training.Propagation.Resilient;
    const NetworkUtil = Encog.Utils.Network;
    const Datasets = Encog.Utils.Datasets;
    const RPROPTypes = ResilientPropagation.getResilientTypes();

    beforeEach(function () {

    });

    describe('Using default Resilient Type (iRPROPp)', function () {
        test('XOR Dataset', function () {
            // train the neural network
            const dataset = Datasets.getXORDataSet();
            const network = NetworkUtil.createXORNetwork();
            const train = new ResilientPropagation(network, dataset.input, dataset.output);

            NetworkUtil.trainNetwork(train);
            const accuracy = NetworkUtil.validateNetwork(network, dataset.input, dataset.output);

            expect(accuracy).toBeGreaterThan(74);
        });

        test('Iris Flower Dataset using normalized data', function () {
            // train the neural network
            const irisDataset = Datasets.getNormalizedIrisDataSet();
            const network = NetworkUtil.createIrisNetwork();

            const train = new ResilientPropagation(network, irisDataset.train.input, irisDataset.train.output);

            NetworkUtil.trainNetwork(train, {minError: 0.01, minIterations: 5});
            const accuracy = NetworkUtil.validateNetwork(network, irisDataset.test.input, irisDataset.test.output);

            expect(accuracy).toBeGreaterThan(70);
        });
    });

    describe('Using RPROP+ type', function () {
        test('Iris Flower Dataset using normalized data with RPROP+ type', function () {
            // train the neural network
            const irisDataset = Datasets.getNormalizedIrisDataSet();
            const network = NetworkUtil.createIrisNetwork();

            const train = new ResilientPropagation(network, irisDataset.train.input, irisDataset.train.output);
            train.setResilientType(RPROPTypes.RPROPp);

            NetworkUtil.trainNetwork(train, {minError: 0.01, minIterations: 5});
            const accuracy = NetworkUtil.validateNetwork(network, irisDataset.test.input, irisDataset.test.output);

            expect(accuracy).toBeGreaterThan(75);
        });
    });

    describe('Using RPROP- type', function () {
        test('Iris Flower Dataset using normalized data with RPROP- type', function () {
            // train the neural network
            const irisDataset = Datasets.getNormalizedIrisDataSet();
            const network = NetworkUtil.createIrisNetwork();

            const train = new ResilientPropagation(network, irisDataset.train.input, irisDataset.train.output);
            train.setResilientType(RPROPTypes.RPROPm);

            NetworkUtil.trainNetwork(train, {minError: 0.01, minIterations: 5});
            const accuracy = NetworkUtil.validateNetwork(network, irisDataset.test.input, irisDataset.test.output);

            expect(accuracy).toBeGreaterThan(75);
        });
    });

    describe('Using iRPROP- type', function () {
        test('Iris Flower Dataset using normalized data with iRPROP- type', function () {
            // train the neural network
            const irisDataset = Datasets.getNormalizedIrisDataSet();
            const network = NetworkUtil.createIrisNetwork();

            const train = new ResilientPropagation(network, irisDataset.train.input, irisDataset.train.output);
            train.setResilientType(RPROPTypes.iRPROPm);

            NetworkUtil.trainNetwork(train, {minError: 0.01, minIterations: 5});
            const accuracy = NetworkUtil.validateNetwork(network, irisDataset.test.input, irisDataset.test.output);

            expect(accuracy).toBeGreaterThan(80);
        });
    });

    describe('Using ARPROP type', function () {
        test('Iris Flower Dataset using normalized data with ARPROP type', function () {
            // train the neural network
            const irisDataset = Datasets.getNormalizedIrisDataSet();
            const network = NetworkUtil.createIrisNetwork();

            const train = new ResilientPropagation(network, irisDataset.train.input, irisDataset.train.output);
            train.setResilientType(RPROPTypes.ARPROP);

            NetworkUtil.trainNetwork(train, {minError: 0.01, minIterations: 5});
            const accuracy = NetworkUtil.validateNetwork(network, irisDataset.test.input, irisDataset.test.output);

            expect(accuracy).toBeGreaterThan(80);
        });
    });
});