neural_network

A neural network library for Dart developers. Build your own networks. Extends the library functions and contribute.

Usage data

To start using artificial neural networks on the data, we treat these data. There are two types of sets, supervised and unsupervised :

// The first parameter is the title of the dataSet.
// The second parameter sets the number of attributes of the dataset.
// The third parameter is optional and indicates the number of class values of the dataset.
// In this case the data set is suspervised because it has class values (2).

DataSet supervisedDataSetExample = new DataSet("SupervisedDataSetTest", 2, numClassValues: 2);

// For an unsupervised data set dont use the optional parameter.

DataSet unsupervisedDataSetExample = new DataSet("UnSupervisedDataSetTest", 2);

Add instances to your data set:

//
// Adding instances to the data set. The number of values in the list must be the same as the number of values and class values.
// If not an exception will be throw.If you add an instance that is already in the dataset, this instance will not be added.
//

supervisedDataSetExample.addInstance([1.0,1.1,1.2,1.3]);
unsupervisedDataSetExample.addInstance([1.0,1.1]);

//
// You can set if the new instance for add will be for train or for test.
//

supervisedDataSetExample.addInstance([0.0,0.1,0.2,0.3], true); // For train
unsupervisedDataSetExample.addInstance([0.0,0.1], false); // For test

//
// Adding multiple instance. All instance must have the same number of values as the number of attributes and class values
// of the data set.
//
supervisedDataSetExample.addInstances([[2.0,2.1,2.2,2.3],[3.0,3.1,3.2,3.3]]);
unsupervisedDataSetExample.addInstances([[2.0,2.1],[3.0,3.1]]);

Removing instances:

  //
  // Removing instances. You can remove one instance or all instance from the dataSet.
  // For removing one instance just provide the index of that instance.
  //

  supervisedDataSetExample.removeInstance(3);
  unsupervisedDataSetExample.removeInstance(2);

  supervisedDataSetExample.removeAllInstances();
  unsupervisedDataSetExample.removeAllInstances();

Its possible to set label for the attributes of the data set:

//
// You can set the labels of the attributes and class values. Be sure that the number of labels is the same as the
// number of attributes and class values. It is not equal an exception will be throw.
//

supervisedDataSetExample.labels = ["First", "Second", "Third", "Fourth"];
unsupervisedDataSetExample.labels = ["First", "Second"];

You must choose which instances of the data set will be used to train and test them for :

//
// Set the number of instance for train and the instance for test.
//

supervisedDataSetExample.trainTestSet = 2; /// The dataset has 3 instances. 2 will be for train and 1 for test.
unsupervisedDataSetExample.trainTestSet = 2; /// The dataset has 3 instances. 2 will be for train and 1 for test.

You may know many parameters of the data set :

  //
  // DataSet info. You can get a lot of information about the dataset.
  //

  supervisedDataSetExample.addInstances([[2.0,2.1,2.2,2.3],[3.0,3.1,3.2,3.3]]);
  unsupervisedDataSetExample.addInstances([[2.0,2.1],[3.0,3.1]]);

  // Get the number of attributes.

  int numAttributesSupervised = supervisedDataSetExample.numValues;
  int numAttributesUnsupervised = unsupervisedDataSetExample.numValues;

  // Get the number of class values. If the number of class values is 0, the dataset will be unsupervised.

  int numClassValuesSupervised = supervisedDataSetExample.numClassValues;
  int numClassValuesUnsupervised = unsupervisedDataSetExample.numClassValues; //Its 0

  // Get if the dataset is supervised. If true, the dataset is supervised, if not the dataset is unsupervised.

  supervisedDataSetExample.isSupervised; //True
  unsupervisedDataSetExample.isSupervised; //False

  // Get the minimum value present in each attribute or class value. If there are no instances it will return an empty array.

  List <double> minValuesSupervised = supervisedDataSetExample.minValues;
  List <double> minValuesUnsupervised = unsupervisedDataSetExample.minValues;

  // Get the maximum value present in each attribute or class value. If there are no instances it will return an empty array.

  List <double> maxValuesSupervised = supervisedDataSetExample.maxValues;
  List <double> maxValuesUnsupervised = unsupervisedDataSetExample.maxValues;

  // Get the mean value in each attribute or class value. If there are no instances it will return an empty array.

  List <double> meanValuesSupervised = supervisedDataSetExample.meanValues;
  List <double> meanValuesUnsupervised = unsupervisedDataSetExample.meanValues;

The application of filters to the data set is also possible :

  // Filters
  //
  // This library constains 2 basic filters that you can apply to a dataset.
  //

  //
  // Normalization filter. If there are no instances it will throw an exception.
  //

  NormalizationFilter normalization = new NormalizationFilter();
  supervisedDataSetExample.instances = normalization.applyFilter(supervisedDataSetExample);
  unsupervisedDataSetExample.instances = normalization.applyFilter(unsupervisedDataSetExample);

  //
  // Randomization filter. If there are no instances it will throw an exception.
  //

  RandomizeFilter random = new RandomizeFilter();
  supervisedDataSetExample.instances = random.applyFilter(supervisedDataSetExample);
  unsupervisedDataSetExample.instances = random.applyFilter(unsupervisedDataSetExample);

There are also four already implemented network topologies :

Adaline:

  // Adaline network.
    // The first parameter determines the number of cells input from the network.
    // The second parameter sets the maximum number of iterations of the network during learning.

    Adaline adalineNetwork = new Adaline(3,100);

    // Learn Process

    // Set the learning rate.

    adalineNetwork.learningRule.learningRate = 0.01;

    // Learn

    adalineNetwork.learningRule.learn(supervisedDataSetExample); //The dataset must be supervised

    // Test your network with the test instance of your dataSet.

    TestingRule testRuleAdaline = new TestingRule(new MeanSquareError(),adalineNetwork);
    testRuleAdaline.test(supervisedDataSetExample);

    List<List<double>>outputNetworkTestAdaline = testRuleAdaline.outputNetworkTest;
    List<double> errorTestAdaline = testRuleAdaline.errorTests;

    // Get the error of all iterations.

    List<double> errorIterationsAdaline = (adalineNetwork.learningRule as BasicLearningRule).errorIterations;

    // Get the outputs of the network in the learn process.

    List<List<double>> outputNetworkTrainAdaline = (adalineNetwork.learningRule as BasicLearningRule).outputNetworkTrain;

Simple Perceptron:

  // Simple Perceptron network.
    // The first parameter determines the number of cells input from the network.
    // The second parameter sets the maximum number of iterations of the network during learning.

    Perceptron simplePerceptron = new Perceptron(3,100);

    // Learn Process

    // Set the learning rate.

    simplePerceptron.learningRule.learningRate = 0.01;

    // Learn

    simplePerceptron.learningRule.learn(supervisedDataSetExample); //The dataset must be supervised. The class value has to ve 1.o for one class and -1 for the other.

    // Get the error of all iterations.

    List<double> errorIterationsPerceptron = (simplePerceptron.learningRule as BasicLearningRule).errorIterations;

    // Get the outputs of the network in the learn process.

    List<List<double>> outputNetworkTrainPerceptron = (simplePerceptron.learningRule as BasicLearningRule).outputNetworkTrain;

    // Test your network with the test instance of your dataSet.

    TestingRule testRuleSimplePerceptron = new TestingRule(new MeanSquareError(),simplePerceptron);
    testRuleSimplePerceptron.test(supervisedDataSetExample);

    List<List<double>>outputNetworkTestSimplePerceptron = testRuleSimplePerceptron.outputNetworkTest;
    List<double> errorTestSimplePerceptron = testRuleSimplePerceptron.errorTests;

Multilayer Perceptron:

    // Multilayer Perceptron
      // The first parameter is a list of integer values. Each value indicates the number of neurons in one layer. The lenght
      // of the list indicates the number of network layers.
      // The second parameter sets the maximum number of iterations of the network during learning.

      MultilayerPerceptron multilayerPerceptron = new MultilayerPerceptron([3,3,3,3,3,1], 100);

      multilayerPerceptron.learningRule.learn(supervisedDataSetExample); //Supervised dataset needed.

      // Get the error of all iterations.

      List<double> errorIterationsMultilayer = (multilayerPerceptron.learningRule as BackPropagationLearningRule).errorIterations;

      // Test your network with the test instance of your dataSet.

      TestingRule testRuleMultilayerPerceptron = new TestingRule(new MeanSquareError(),multilayerPerceptron);
      testRuleMultilayerPerceptron.test(supervisedDataSetExample);

      List<List<double>>outputNetworkTestMultilayerPerceptron = testRuleMultilayerPerceptron.outputNetworkTest;
      List<double> errorTestMultilayerPerceptron = testRuleMultilayerPerceptron.errorTests;

Radial Base network:

    // Radial base network
    // The first parameter defines the number of neurons in the input layer .
    // The second parameter defines the number of neurons in the hidden layer.
    // The third parameter refers to the number of neurons in the output layer.
    // The last parameter is the maximum number of iterations for learning.

    RadialBase radialNetwork = new RadialBase(8,3,1,500);

    //An initialization of the network its needed.

    List<List<double>> trainInstances = [];
    for(int i = 0; i < dataSetTest.instances.length; i++){
      if(dataSetTest.instances[i].isForTrain!=null && dataSetTest.instances[i].isForTrain)
        trainInstances.add(dataSetTest.instanceValues(i));
    }

    (radialNetwork.learningRule as RadialLearning).initialization(trainInstances);

    //Learn

    radialNetwork.learningRule.learningRate = 0.0001;
    radialNetwork.learningRule.learn(dataSetTest);

    // Test

    TestingRule testRule = new TestingRule(new MeanSquareError(),radialNetwork);
    testRule.test(dataSetTest);