nlp-toolkit/lib/calculate/derivation.js

This module covers some basic nlp principles and implementations. Every implementation in this module is written as stream to only hold that data in memory that is currently processed at any step.

/**
 * Calculate standard derivation.
 */


'use strict';


/**
 * MODULES.
 */
var math = require('mathjs');


/**
 * FUNCTIONS.
 */
function derivation(data, options) {
  var derivs = options.derivation || 1;
  return data.map(function (classifier) {
    var results = {
      lost: 0,
      positives: 0,
      negatives: 0,
      won: 0,
      positivesWon: 0,
      negativesWon: 0
    };
    classifier.forEach(function (_set) {
      _set.forEach(function (entry) {
        var guess = entry.guess.category || entry.guess;
        entry.guess.distribution.sort(function (a, b) {
          return (a.prob > b.prob) ? -1 : 1;
        });
        var _deriv = math.std(entry.guess.distribution.map(function (cat) {
          return cat.prob;
        }));
        if (entry.guess.distribution[0].prob - (_deriv * derivs) <= entry.guess.distribution[1].prob) {
          results.lost++;
          if (guess === entry.feature) {
            results.positives++;
          } else {
            results.negatives++;
          }
        } else {
          results.won++;
          if (guess === entry.feature) {
            results.positivesWon++;
          } else {
            results.negativesWon++;
          }
        }
      });
    });
    results.positives = results.positives / results.lost;
    results.negatives = results.negatives / results.lost;
    results.positivesWon = results.positivesWon / results.won;
    results.negativesWon = results.negativesWon / results.won;
    return results;
  });
}


/**
 * EXPORTS.
 */
module.exports = derivation;