@jsmlt/jsmlt/distribution/supervised/linear/perceptron.js

JavaScript Machine Learning

'use strict';

Object.defineProperty(exports, "__esModule", {
  value: true
});
exports.BinaryPerceptron = undefined;

var _extends = Object.assign || function (target) { for (var i = 1; i < arguments.length; i++) { var source = arguments[i]; for (var key in source) { if (Object.prototype.hasOwnProperty.call(source, key)) { target[key] = source[key]; } } } return target; };

var _slicedToArray = function () { function sliceIterator(arr, i) { var _arr = []; var _n = true; var _d = false; var _e = undefined; try { for (var _i = arr[Symbol.iterator](), _s; !(_n = (_s = _i.next()).done); _n = true) { _arr.push(_s.value); if (i && _arr.length === i) break; } } catch (err) { _d = true; _e = err; } finally { try { if (!_n && _i["return"]) _i["return"](); } finally { if (_d) throw _e; } } return _arr; } return function (arr, i) { if (Array.isArray(arr)) { return arr; } else if (Symbol.iterator in Object(arr)) { return sliceIterator(arr, i); } else { throw new TypeError("Invalid attempt to destructure non-iterable instance"); } }; }();

var _createClass = function () { function defineProperties(target, props) { for (var i = 0; i < props.length; i++) { var descriptor = props[i]; descriptor.enumerable = descriptor.enumerable || false; descriptor.configurable = true; if ("value" in descriptor) descriptor.writable = true; Object.defineProperty(target, descriptor.key, descriptor); } } return function (Constructor, protoProps, staticProps) { if (protoProps) defineProperties(Constructor.prototype, protoProps); if (staticProps) defineProperties(Constructor, staticProps); return Constructor; }; }();

var _base = require('../base');

var _linalg = require('../../math/linalg');

var LinAlg = _interopRequireWildcard(_linalg);

function _interopRequireWildcard(obj) { if (obj && obj.__esModule) { return obj; } else { var newObj = {}; if (obj != null) { for (var key in obj) { if (Object.prototype.hasOwnProperty.call(obj, key)) newObj[key] = obj[key]; } } newObj.default = obj; return newObj; } }

function _classCallCheck(instance, Constructor) { if (!(instance instanceof Constructor)) { throw new TypeError("Cannot call a class as a function"); } }

function _possibleConstructorReturn(self, call) { if (!self) { throw new ReferenceError("this hasn't been initialised - super() hasn't been called"); } return call && (typeof call === "object" || typeof call === "function") ? call : self; }

function _inherits(subClass, superClass) { if (typeof superClass !== "function" && superClass !== null) { throw new TypeError("Super expression must either be null or a function, not " + typeof superClass); } subClass.prototype = Object.create(superClass && superClass.prototype, { constructor: { value: subClass, enumerable: false, writable: true, configurable: true } }); if (superClass) Object.setPrototypeOf ? Object.setPrototypeOf(subClass, superClass) : subClass.__proto__ = superClass; } // Internal dependencies


/**
 * Perceptron learner for binary classification problem.
 */
var BinaryPerceptron = exports.BinaryPerceptron = function (_Classifier) {
  _inherits(BinaryPerceptron, _Classifier);

  function BinaryPerceptron() {
    _classCallCheck(this, BinaryPerceptron);

    return _possibleConstructorReturn(this, (BinaryPerceptron.__proto__ || Object.getPrototypeOf(BinaryPerceptron)).apply(this, arguments));
  }

  _createClass(BinaryPerceptron, [{
    key: 'getClassIndexSign',

    /**
     * Get the signed value of the class index. Returns -1 for class index 0, 1 for class index 1.
     *
     * @param {number} classIndex - Class index
     * @return {number} Sign corresponding to class index
     */
    value: function getClassIndexSign(classIndex) {
      return classIndex * 2 - 1;
    }

    /**
     * Get the class index corresponding to a sign.
     *
     * @param {number} sign - Sign
     * @return {number} Class index corresponding to sign
     */

  }, {
    key: 'getSignClassIndex',
    value: function getSignClassIndex(sign) {
      return (sign + 1) / 2;
    }

    /**
     * @see {Classifier#train}
     */

  }, {
    key: 'train',
    value: function train(X, y) {
      // Weights increment to check for convergence
      this.weightsIncrement = Infinity;

      // Initialize weights vector to zero. Here, the number of weights equals one plus the number of
      // features, where the first weight (w0) is the weight used for the bias.
      this.weights = LinAlg.zeroVector(1 + X[0].length);

      // Store historic errors
      var epochNumErrors = [];

      // Iteration index
      var epoch = 0;

      // A single iteration of this loop corresponds to a single iteration of training all data
      // points in the data set
      while (true) {
        var _trainIteration = this.trainIteration(X, y),
            _trainIteration2 = _slicedToArray(_trainIteration, 2),
            numErrors = _trainIteration2[0],
            weightsIncrement = _trainIteration2[1];

        epochNumErrors.push(numErrors);

        if (weightsIncrement.reduce(function (r, a) {
          return r + Math.abs(a);
        }, 0) < 0.0001 || epoch > 100) {
          break;
        }

        epoch += 1;
      }
    }

    /**
     * Train the classifier for a single iteration on the stored training data.
     *
     * @param {Array.<Array.<number>>} X - Features per data point
     * @param {Array.<mixed>} y Class labels per data point
     */

  }, {
    key: 'trainIteration',
    value: function trainIteration(X, y) {
      // Initialize the weights increment vector, which is used to increment the weights in each
      // iteration after the calculations are done.
      var weightsIncrement = LinAlg.zeroVector(this.weights.length);

      // Initialize number of misclassified points
      var numErrors = 0;

      // Shuffle data points

      var _LinAlg$permuteRows = LinAlg.permuteRows(X, y),
          _LinAlg$permuteRows2 = _slicedToArray(_LinAlg$permuteRows, 2),
          XUse = _LinAlg$permuteRows2[0],
          yUse = _LinAlg$permuteRows2[1];

      // Loop over all datapoints


      for (var i = 0; i < XUse.length; i += 1) {
        // Transform binary class index to class sign (0 becomes -1 and 1 remains 1)
        var classSign = this.getClassIndexSign(yUse[i]);

        // Copy features vector so it is not changed in the datapoint
        var augmentedFeatures = XUse[i].slice();

        // Add feature with value 1 at the beginning of the feature vector to correpond with the
        // bias weight
        augmentedFeatures.unshift(1);

        // Calculate output
        var output = LinAlg.dot(augmentedFeatures, this.weights);

        // Check whether the point was correctly classified
        if (classSign * output <= 0) {
          // Increase the number of errors
          numErrors += 1;

          // Update the weights change to be used at the end of this epoch
          weightsIncrement = LinAlg.sum(weightsIncrement, LinAlg.scale(augmentedFeatures, classSign));
        }
      }

      // Take average of all weight increments
      this.weightsIncrement = LinAlg.scale(weightsIncrement, 0.01 / XUse.length);
      this.weights = LinAlg.sum(this.weights, this.weightsIncrement);

      return [numErrors, weightsIncrement];
    }

    /**
     * Check whether training has convergence when using iterative training using trainIteration.
     *
     * @return {boolean} Whether the algorithm has converged
     */

  }, {
    key: 'checkConvergence',
    value: function checkConvergence() {
      return LinAlg.internalSum(LinAlg.abs(this.weightsIncrement)) < 0.0001;
    }

    /**
     * Make a prediction for a data set.
     *
     * @param {Array.Array.<number>} features - Features for each data point
     * @param {Object} [optionsUser] User-defined options
     * @param {string} [optionsUser.output = 'classLabels'] Output for predictions. Either
     *   "classLabels" (default, output predicted class label), "raw" (dot product of weights vector
     *   with augmented features vector) or "normalized" (dot product from "raw" but with unit-length
     *   weights)
     * @return {Array.<number>} Predictions. Output dependent on options.output, defaults to class
     *   labels
     */

  }, {
    key: 'predict',
    value: function predict(features) {
      var optionsUser = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};

      // Options
      var optionsDefault = {
        output: 'classLabels' // 'classLabels', 'normalized' or 'raw'
      };

      var options = _extends({}, optionsDefault, optionsUser);

      // Predictions
      var predictions = [];

      // Normalization factor for normalized output
      var weightsMagnitude = Math.sqrt(LinAlg.dot(this.weights, this.weights));

      // Loop over all datapoints
      for (var i = 0; i < features.length; i += 1) {
        // Copy features vector so it is not changed in the datapoint
        var augmentedFeatures = features[i].slice();

        // Add feature with value 1 at the beginning of the feature vector to correpond with the
        // bias weight
        augmentedFeatures.unshift(1);

        // Calculate output
        var output = LinAlg.dot(augmentedFeatures, this.weights);

        // Store prediction
        if (options.output === 'raw') {
          // Raw output: do nothing
        } else if (options.output === 'normalized') {
          // Normalized output
          output /= weightsMagnitude;
        } else {
          // Class label output
          output = this.getSignClassIndex(output > 0 ? 1 : -1);
        }

        predictions.push(output);
      }

      return predictions;
    }
  }]);

  return BinaryPerceptron;
}(_base.Classifier);

/**
 * Perceptron learner for 2 or more classes. Uses 1-vs-all classification.
 */


var Perceptron = function (_OneVsAllClassifier) {
  _inherits(Perceptron, _OneVsAllClassifier);

  /**
   * Constructor. Initialize class members and store user-defined options.
   *
   * @param {Object} [optionsUser] User-defined options
   * @param {trackAccuracy} [optionsUser.trackAccuracy = false] Whether to track accuracy during the
   *   training process. This will let the perceptron keep track of the error rate on the test set
   *   in each training iteration
   */
  function Perceptron() {
    var optionsUser = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : {};

    _classCallCheck(this, Perceptron);

    // Parse options
    var _this2 = _possibleConstructorReturn(this, (Perceptron.__proto__ || Object.getPrototypeOf(Perceptron)).call(this));

    var optionsDefault = {
      // Whether the number of misclassified samples should be tracked at each iteration
      trackAccuracy: false
    };

    var options = _extends({}, optionsDefault, optionsUser);

    // Set options
    _this2.trackAccuracy = options.trackAccuracy;

    // Accuracy tracking
    if (_this2.trackAccuracy) {
      _this2.addListener('iterationCompleted', function () {
        return _this2.calculateIntermediateAccuracy();
      });
    }
    return _this2;
  }

  /**
   * @see {@link OneVsAll#createClassifier}
   */


  _createClass(Perceptron, [{
    key: 'createClassifier',
    value: function createClassifier(classIndex) {
      return new BinaryPerceptron();
    }

    /**
     * @see {@link Classifier#train}
     */

  }, {
    key: 'train',
    value: function train(X, y) {
      this.createClassifiers(y);

      if (this.trackAccuracy) {
        this.numErrors = [];
        this.trainIterative();
      } else {
        this.trainBatch(X, y);
      }
    }

    /**
     * Callback for calculating the accuracy of the classifier on the training set in intermediate
     * steps of training
     */

  }, {
    key: 'calculateIntermediateAccuracy',
    value: function calculateIntermediateAccuracy() {
      var _this3 = this;

      // Track number of errors
      var predictions = this.predict(this.training.features);
      this.numErrors.push(predictions.reduce(function (r, x, i) {
        return r + (x !== _this3.training.labels[i]);
      }, 0));
    }
  }]);

  return Perceptron;
}(_base.OneVsAllClassifier);

exports.default = Perceptron;