gaussianMixture - npm Package Compare versions

Comparing version 0.1.0 to 0.2.0

index.js

@@ -1,14 +0,14 @@
-'use strict'
+'use strict';
 
 // Imports
-var gaussian = require('gaussian')
-var _ = require('underscore')
+var gaussian = require('gaussian');
+var _ = require('underscore');
 
 // Constants
-var MAX_ITERATIONS = 200
-var LOG_LIKELIHOOD_TOL = 1e-7
+var MAX_ITERATIONS = 200;
+var LOG_LIKELIHOOD_TOL = 1e-7;
 
-module.exports = GMM
+module.exports = GMM;
 
-function GMM (nComponents, weights, means, vars) {
+function GMM(nComponents, weights, means, vars, options) {
   // nComponents: number of components in the mixture
@@ -18,20 +18,21 @@ // weights: array of weights of each component in the mixture, must sum to 1
   // vars: array of variances of each component
-  this.nComponents = nComponents
-  this.weights = weights === undefined ? _.range(nComponents).map(function () { return 1 / nComponents }) : weights
-  this.means = means === undefined ? _.range(nComponents) : means
-  this.vars = vars === undefined ? _.range(nComponents).map(function () { return 1 }) : vars
+  this.nComponents = nComponents;
+  this.weights = weights === undefined ? _.range(nComponents).map(function () { return 1 / nComponents; }) : weights;
+  this.means = means === undefined ? _.range(nComponents) : means;
+  this.vars = vars === undefined ? _.range(nComponents).map(function () { return 1; }) : vars;
   if (nComponents !== this.weights.length ||
-    nComponents !== this.means.length ||
-    nComponents !== this.vars.length) {
-    throw new Error('weights, means and vars must have nComponents elements.')
+      nComponents !== this.means.length ||
+      nComponents !== this.vars.length) {
+    throw new Error('weights, means and vars must have nComponents elements.');
   }
+  this.options = options === undefined ? {} : options;
 }
 
 GMM.prototype._gaussians = function () {
-  var gaussians = []
+  var gaussians = [];
   for (var k = 0; k < this.nComponents; k++) {
-    gaussians.push(gaussian(this.means[k], this.vars[k]))
+    gaussians.push(gaussian(this.means[k], this.vars[k]));
   }
-  return gaussians
-}
+  return gaussians;
+};
 
@@ -41,16 +42,16 @@ GMM.prototype.sample = function (nSamples) {
   // return: array of length nSamples
-  var samples = []
-  var gaussians = this._gaussians()
+  var samples = [];
+  var gaussians = this._gaussians();
   // generate samples
   for (var i = 0; i < nSamples; i++) {
-    var r = Math.random()
-    var n = 0
+    var r = Math.random();
+    var n = 0;
     while (r > this.weights[n] && n < this.nComponents) {
-      r -= this.weights[n]
-      n++
+      r -= this.weights[n];
+      n++;
     }
-    samples.push(gaussians[n].ppf(Math.random()))
+    samples.push(gaussians[n].ppf(Math.random()));
   }
-  return samples
-}
+  return samples;
+};
 
@@ -60,8 +61,8 @@ GMM.prototype.memberships = function (data) {
   // return: (data.length * this.nComponents) matrix with membership weights
-  var memberships = []
+  var memberships = [];
   for (var i = 0, n = data.length; i < n; i++) {
-    memberships.push(this.membership(data[i]))
+    memberships.push(this.membership(data[i]));
   }
-  return memberships
-}
+  return memberships;
+};
 
@@ -71,10 +72,10 @@ GMM.prototype.membership = function (x) {
   // return: array of probabilities of length this.nComponents
-  var membership = []
-  var gaussians = this._gaussians()
+  var membership = [];
+  var gaussians = this._gaussians();
   for (var i = 0; i < this.nComponents; i++) {
-    membership.push(gaussians[i].pdf(x))
+    membership.push(gaussians[i].pdf(x));
   }
-  var sum = membership.reduce(function (a, b) { return a + b })
-  return membership.map(function (a) { return a / sum })
-}
+  var sum = membership.reduce(function (a, b) { return a + b; });
+  return membership.map(function (a) { return a / sum; });
+};
 
@@ -84,21 +85,24 @@ GMM.prototype.updateModel = function (data) {
   // It will update the GMM weights, means and variances.
+  // Optionally, if options.variancePrior and options.priorRelevance are defined,
+  // mix in the prior.
 
   // First, we compute the data memberships.
-  var n = data.length
-  var memberships = this.memberships(data)
+  var n = data.length;
+  var memberships = this.memberships(data);
 
   // Update the mixture weights
-  var componentWeights = []
+  var componentWeights = [];
+  var reduceFunction = function (k) { return function (a, b) { return (a + b[k]); }; };
   for (var k = 0; k < this.nComponents; k++) {
-    componentWeights[k] = memberships.reduce(function (a, b) { return (a + b[k]) }, 0)
+    componentWeights[k] = memberships.reduce(reduceFunction(k), 0);
   }
-  this.weights = componentWeights.map(function (a) { return a / n })
+  this.weights = componentWeights.map(function (a) { return a / n; });
 
   // Update the mixture means
   for (k = 0; k < this.nComponents; k++) {
-    this.means[k] = 0
+    this.means[k] = 0;
     for (var i = 0; i < n; i++) {
-      this.means[k] += memberships[i][k] * data[i]
+      this.means[k] += memberships[i][k] * data[i];
     }
-    this.means[k] /= componentWeights[k]
+    this.means[k] /= componentWeights[k];
   }
@@ -108,39 +112,44 @@
   for (k = 0; k < this.nComponents; k++) {
-    this.vars[k] = 0
+    this.vars[k] = 0;
     for (i = 0; i < n; i++) {
-      this.vars[k] += memberships[i][k] * Math.pow(data[i] - this.means[k], 2)
+      this.vars[k] += memberships[i][k] * Math.pow(data[i] - this.means[k], 2);
     }
-    this.vars[k] /= componentWeights[k]
+    this.vars[k] /= componentWeights[k];
+    // If there is a variance prior:
+    if (this.options.variancePrior && this.options.priorRelevance) {
+      var alpha = this.weights[k] / (this.weights[k] + this.options.priorRelevance);
+      this.vars[k] = alpha * this.vars[k] + (1 - alpha) * this.options.variancePrior;
+    }
   }
-}
+};
 
 GMM.prototype.logLikelihood = function (data) {
   // data: array of floats representing the samples to compute the log-likelihood with.
-  var l = 0
-  var p = 0
-  var gaussians = this._gaussians()
+  var l = 0;
+  var p = 0;
+  var gaussians = this._gaussians();
   for (var i = 0, n = data.length; i < n; i++) {
-    p = 0
+    p = 0;
     for (var k = 0; k < this.nComponents; k++) {
-      p += this.weights[k] * gaussians[k].pdf(data[i])
+      p += this.weights[k] * gaussians[k].pdf(data[i]);
     }
-    l += Math.log(p)
+    l += Math.log(p);
   }
-  return l
-}
+  return l;
+};
 
 GMM.prototype.optimize = function (data, maxIterations, logLikelihoodTol) {
   // data: array of floats
-  maxIterations = maxIterations === undefined ? MAX_ITERATIONS : maxIterations
-  logLikelihoodTol = logLikelihoodTol === undefined ? LOG_LIKELIHOOD_TOL : logLikelihoodTol
-  var logLikelihoodDiff = Infinity
-  var logLikelihood = -Infinity
-  var temp
+  maxIterations = maxIterations === undefined ? MAX_ITERATIONS : maxIterations;
+  logLikelihoodTol = logLikelihoodTol === undefined ? LOG_LIKELIHOOD_TOL : logLikelihoodTol;
+  var logLikelihoodDiff = Infinity;
+  var logLikelihood = -Infinity;
+  var temp;
   for (var i = 0; i < maxIterations && logLikelihoodDiff > logLikelihoodTol; i++) {
-    this.updateModel(data)
-    temp = this.logLikelihood(data)
-    logLikelihoodDiff = Math.abs(logLikelihood - temp)
-    logLikelihood = temp
+    this.updateModel(data);
+    temp = this.logLikelihood(data);
+    logLikelihoodDiff = Math.abs(logLikelihood - temp);
+    logLikelihood = temp;
   }
-  return i // number of steps to reach the converged solution.
-}
+  return i; // number of steps to reach the converged solution.
+};

package.json

 {
   "name": "gaussianMixture",
-  "version": "0.1.0",
+  "version": "0.2.0",
   "description": "An implementation of a Gaussian Mixture class in one dimension, that allows to fit models with an Expectation Maximization algorithm.",
@@ -8,3 +8,5 @@ "main": "index.js",
   "scripts": {
-    "test": "standard && tap test/*.test.js",
+    "test": "npm run lint && tap test/*.test.js",
+    "lint": "eslint '*.js' 'test/*.test.js'",
+    "fix": "eslint --fix '*.js' 'test/*.test.js'",
     "cover": "tap test/*.test.js --cov --coverage-report=lcov"
@@ -21,3 +23,4 @@ },
   "devDependencies": {
-    "standard": "~7.1.2",
+    "eslint": "^3.4.0",
+    "eslint-config-mourner": "^2.0.1",
     "tap": "~6.2.0"
@@ -24,0 +27,0 @@ },

README.md

@@ -21,2 +21,20 @@ # Gaussian Mixture
 gmm.optimize(data) // updates weights, means and variances with the EM algorithm given the data.
+console.log(gmm.means); // >> [1.225, 7.3, 14.8]
 ```
+
+## Options
+
+Optionally, define an options object and pass it at instantiation:
+
+```
+var options = {
+  variancePrior: // Float,
+  priorRelevance: // Positive float.
+};
+
+var gmm = new GMM(nComponents, weights, means, vars, options);
+```
+
+The variance prior allows you to define a prior for the variance of all components (assumed the same). This prior will be mixed in the maximization step of the EM optimization with a relevance score equal to `options.priorRelevance`.
+
+The mixing weight `alpha` for component `i` is `alpha = weights[i] / (weights[i] + options.priorRelevance)`.

test/gmm.test.js

@@ -1,103 +0,132 @@
-var test = require('tap').test
-var GMM = require('../index')
+'use strict';
 
+var test = require('tap').test;
+var data = require('./fixtures/data.json'); // 200 samples from refGmm
+
+var GMM = require('../index');
+
 test('Initialization of a new GMM object.', function (t) {
-  t.plan(2)
+  t.plan(2);
 
-  function f (k) {
+  function f(k) {
     return function () {
-      var gmm = new GMM(3, k)
-      gmm.sample(0)
-    }
+      var gmm = new GMM(3, k);
+      gmm.sample(0);
+    };
   }
-  t.throws(f([1, 2]))
-  t.doesNotThrow(f([1, 2, 3]))
-})
+  t.throws(f([1, 2]));
+  t.doesNotThrow(f([1, 2, 3]));
+});
 
 test('Random sampling.', function (t) {
-  t.plan(1)
+  t.plan(1);
 
-  var gmm = new GMM(3)
-  t.equals(5, gmm.sample(5).length)
-})
+  var gmm = new GMM(3);
+  t.equals(5, gmm.sample(5).length);
+});
 
 test('Gaussians of a mixture model.', function (t) {
-  t.plan(6)
+  t.plan(6);
 
-  var gmm = new GMM(3, [1 / 3, 1 / 3, 1 / 3], [0, 10, 20], [1, 2, 0.5])
-  var gaussians = gmm._gaussians()
-  t.equals(gaussians[0].mean, 0)
-  t.equals(gaussians[1].mean, 10)
-  t.equals(gaussians[2].mean, 20)
-  t.equals(gaussians[0].variance, 1)
-  t.equals(gaussians[1].variance, 2)
-  t.equals(gaussians[2].variance, 0.5)
-})
+  var gmm = new GMM(3, [1 / 3, 1 / 3, 1 / 3], [0, 10, 20], [1, 2, 0.5]);
+  var gaussians = gmm._gaussians();
+  t.equals(gaussians[0].mean, 0);
+  t.equals(gaussians[1].mean, 10);
+  t.equals(gaussians[2].mean, 20);
+  t.equals(gaussians[0].variance, 1);
+  t.equals(gaussians[1].variance, 2);
+  t.equals(gaussians[2].variance, 0.5);
+});
 
 test('Computing membership for one datapoint', function (t) {
-  t.plan(2)
+  t.plan(2);
 
-  var gmm = new GMM(3, undefined, [0, 10, 20])
-  t.equals(gmm.membership(5)[0], gmm.membership(5)[1])
-  t.equals(gmm.membership(0)[0] > 0.99, true)
-})
+  var gmm = new GMM(3, undefined, [0, 10, 20]);
+  t.equals(gmm.membership(5)[0], gmm.membership(5)[1]);
+  t.equals(gmm.membership(0)[0] > 0.99, true);
+});
 
 test('Shape of the membership matrix', function (t) {
-  t.plan(2)
+  t.plan(2);
 
-  var gmm = new GMM(5)
-  var memberships = gmm.memberships([1, 2, 3, 4, 5, 6])
-  t.equals(memberships.length, 6)
-  t.equals(memberships[0].length, 5)
-})
+  var gmm = new GMM(5);
+  var memberships = gmm.memberships([1, 2, 3, 4, 5, 6]);
+  t.equals(memberships.length, 6);
+  t.equals(memberships[0].length, 5);
+});
 
 test('Convergence of model update', function (t) {
-  t.plan(9)
+  t.plan(9);
 
-  var refGmm = new GMM(3, [0.2, 0.5, 0.3], [0, 10, 30], [1, 2, 4])
-  var testGmm = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1])
-  var data = require('./fixtures/data.json') // 200 samples from refGmm
+  var refGmm = new GMM(3, [0.2, 0.5, 0.3], [0, 10, 30], [1, 2, 4]);
+  var testGmm = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1]);
   for (var i = 0; i < 200; i++) {
-    testGmm.updateModel(data)
+    testGmm.updateModel(data);
   }
   for (var j = 0; j < 3; j++) {
-    t.equals(Math.abs(testGmm.weights[j] - refGmm.weights[j]) < 0.1, true)
-    t.equals(Math.abs(testGmm.means[j] - refGmm.means[j]) < 1, true)
-    t.equals(Math.abs(testGmm.vars[j] - refGmm.vars[j]) < 1, true)
+    t.equals(Math.abs(testGmm.weights[j] - refGmm.weights[j]) < 0.1, true);
+    t.equals(Math.abs(testGmm.means[j] - refGmm.means[j]) < 1, true);
+    t.equals(Math.abs(testGmm.vars[j] - refGmm.vars[j]) < 1, true);
   }
-})
+});
 
 test('log likelihood function', function (t) {
-  t.plan(15)
+  t.plan(15);
 
-  var gmm = new GMM(3, undefined, [-1, 15, 37], [1, 1, 1])
-  var data = require('./fixtures/data.json') // 200 samples from another GMM
-  var l = -Infinity
-  var temp
+  var gmm = new GMM(3, undefined, [-1, 15, 37], [1, 1, 1]);
+  var l = -Infinity;
+  var temp;
   for (var i = 0; i < 15; i++) {
-    gmm.updateModel(data)
-    temp = gmm.logLikelihood(data)
-    t.equals(temp - l >= -1e-5, true)
-    l = temp
+    gmm.updateModel(data);
+    temp = gmm.logLikelihood(data);
+    t.equals(temp - l >= -1e-5, true);
+    l = temp;
   }
-})
+});
 
 test('EM full optimization', function (t) {
-  t.plan(10)
+  t.plan(10);
 
-  var gmm = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1])
-  var gmm2 = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1])
+  var gmm = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1]);
+  var gmm2 = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1]);
 
-  var data = require('./fixtures/data.json') // 200 samples from refGmm
   for (var i = 0; i < 200; i++) {
-    gmm.updateModel(data)
+    gmm.updateModel(data);
   }
-  var counter = gmm2.optimize(data)
+  var counter = gmm2.optimize(data);
 
-  t.equals(counter, 3)
+  t.equals(counter, 3);
   for (var j = 0; j < 3; j++) {
-    t.equals(Math.abs(gmm.weights[j] - gmm2.weights[j]) < 1e-7, true)
-    t.equals(Math.abs(gmm.means[j] - gmm2.means[j]) < 1e-7, true)
-    t.equals(Math.abs(gmm.vars[j] - gmm2.vars[j]) < 1e-7, true)
+    t.equals(Math.abs(gmm.weights[j] - gmm2.weights[j]) < 1e-7, true);
+    t.equals(Math.abs(gmm.means[j] - gmm2.means[j]) < 1e-7, true);
+    t.equals(Math.abs(gmm.vars[j] - gmm2.vars[j]) < 1e-7, true);
   }
-})
+});
+
+test('Variance prior', function (t) {
+  t.plan(3);
+
+  var options = {
+    variancePrior: 3,
+    priorRelevance: 0.01
+  };
+  var options2 = {
+    variancePrior: 3,
+    priorRelevance: 1
+  };
+  var options3 = {
+    variancePrior: 3,
+    priorRelevance: 1000000
+  };
+  var gmm = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1], options);
+  var gmm2 = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1], options2);
+  var gmm3 = new GMM(3, undefined, [-1, 13, 25], [1, 1, 1], options3);
+  gmm.optimize(data);
+  gmm2.optimize(data);
+  gmm3.optimize(data);
+
+  var cropFloat = function (a) { return Number(a.toFixed(1)); };
+  t.same(gmm.vars.map(cropFloat), [1.1, 2.5, 4.6]);
+  t.same(gmm2.vars.map(cropFloat), [2.7, 2.8, 3.4]);
+  t.same(gmm3.vars.map(cropFloat), [3, 3, 3]);
+});

New alerts

Major refactor

Supply chain risk

Package has recently undergone a major refactor. It may be unstable or indicate significant internal changes. Use caution when updating to versions that include significant changes.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

15713

increased by15.64%

Number of package files

9

increased by12.5%

Lines of code

243

increased by15.71%

Number of lines in readme file

40

increased by81.82%

Worsened metrics

Dev dependency count

3

increased by50%

Number of medium supply chain risk alerts

1

increased byInfinity%

No dependency changes