fix!: algorithm for oneOfWeighted has better distribution (#47)

Fixes #46

Co-authored-by: V.Beyer <v.beyer@unimed.de>
Co-authored-by: Justin van der Merwe <justinvderm@gmail.com>

Author: ctholho

oneOfWeighted.js

@@ -1,7 +1,5 @@
 var hash = require('./hash')
 var hash2 = hash.hash2
-var flip = require('./utils/flip')
-var resolve = require('./utils/resolve')
 
 var EPS = 0.0001
 
@@ -12,19 +10,13 @@ function oneOfWeighted(a, b) {
 function oneOfWeightedMain(input, samples) {
   samples = parseSamples(samples)
   var id = hash2(input, 'oneOfWeighted')
-  var n = samples.length
-  var pRemaining = 1
-  var i = -1
-  var sample
-  var p
-
-  while (++i < n) {
-    sample = samples[i]
-    p = sample[0] / pRemaining
-    pRemaining -= p
+  var prob = (id % 1000000) / 1000000
 
-    if (flip(id, p)) {
-      return resolve(id, sample[1])
+  var cumulative = 0
+  for (var i = 0; i < samples.length; i++) {
+    cumulative += samples[i][0]
+    if (prob < cumulative) {
+      return samples[i][1]
     }
   }
 

readme.md

@@ -800,7 +800,7 @@ oneOfWeighted('id-2', [
   [0.05, char],
   [0.05, int]
 ])
-// => 'Ut'
+// => [Function: word] { options: [Function: wordOptions] }
 ```
 
 For each `[probability, value]` pair in the array of `values`, if the given
@@ -817,7 +817,7 @@ oneOfWeighted('id-23', [
   [null, 'green'],
   [null, 'blue']
 ])
-// => 'green'
+// => 'blue'
 ```
 
 ## <a name="install-use" href="#install-use">#</a> Install & Use

tests/oneOfWeighted.test.js

@@ -4,29 +4,156 @@ const { diffBetween } = require('./utils')
 
 const DIFF_THRESHOLD = 0.065
 
-test(`averages to within ${
-  DIFF_THRESHOLD * 100
-}% of the given probabilities`, t => {
+test(`1/5 averages to within ${DIFF_THRESHOLD * 100
+  }% of the given probabilities`, t => {
+    const n = 10000
+    let i = -1
+
+    const sums = {
+      red: 0,
+      green: 0,
+      blue: 0,
+      yellow: 0,
+      fuchsia: 0
+    }
+
+    const fn = oneOfWeighted([
+      [0.6, 'red'],
+      [0.2, 'green'],
+      [0.1, 'blue'],
+      [0.05, 'yellow'],
+      [0.05, 'fuchsia']
+    ])
+
+    while (++i < n) sums[fn(i)]++
+
+    t.assert(diffBetween(sums.red / n, 0.6) <= DIFF_THRESHOLD)
+    t.assert(diffBetween(sums.green / n, 0.2) <= DIFF_THRESHOLD)
+    t.assert(diffBetween(sums.blue / n, 0.1) <= DIFF_THRESHOLD)
+    t.assert(diffBetween(sums.yellow / n, 0.05) <= DIFF_THRESHOLD)
+    t.assert(diffBetween(sums.fuchsia / n, 0.05) <= DIFF_THRESHOLD)
+  })
+
+test(`2/5 averages to within ${DIFF_THRESHOLD * 100}% of the given probabilities`, t => {
+  const n = 10000
+  let i = -1
+
+  const sums = {
+    apple: 0,
+    banana: 0,
+    cherry: 0,
+    pear: 0,
+    persimmon: 0
+  }
+
+  const fn = oneOfWeighted([
+    [0.4, 'apple'],
+    [0.3, 'banana'],
+    [0.2, 'cherry'],
+    [0.05, 'pear'],
+    [0.05, 'persimmon']
+  ])
+
+  while (++i < n) sums[fn(i)]++
+
+  t.assert(diffBetween(sums.apple / n, 0.4) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.banana / n, 0.3) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.cherry / n, 0.2) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.pear / n, 0.05) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.persimmon / n, 0.05) <= DIFF_THRESHOLD)
+})
+
+test(`3/5 averages to within ${DIFF_THRESHOLD * 100}% of the given probabilities`, t => {
+  const n = 10000
+  let i = -1
+
+  const sums = {
+    Hello: 0,
+    'Ni hao': 0,
+    Namaste: 0,
+    Hola: 0,
+    Bonjour: 0,
+    Salam: 0,
+    Nomoshkar: 0
+  }
+
+  const fn = oneOfWeighted([
+    [0.25, 'Namaste'],
+    [0.2, 'Hello'],
+    [0.2, 'Ni hao'],
+    [0.15, 'Hola'],
+    [0.07, 'Bonjour'],
+    [0.07, 'Salam'],
+    [0.06, 'Nomoshkar']
+  ])
+
+  while (++i < n) sums[fn(i)]++
+
+  t.assert(diffBetween(sums['Namaste'] / n, 0.25) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums['Hello'] / n, 0.2) <= DIFF_THRESHOLD + 0.025)
+  t.assert(diffBetween(sums['Ni hao'] / n, 0.2) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums['Hola'] / n, 0.15) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums['Bonjour'] / n, 0.07) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums['Salam'] / n, 0.07) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums['Nomoshkar'] / n, 0.06) <= DIFF_THRESHOLD)
+})
+
+test(`4/5 averages to within ${DIFF_THRESHOLD * 100}% of the given probabilities`, t => {
+  const n = 10000
+  let i = -1
+
+  const sums = {
+    cat: 0,
+    dog: 0,
+    bird: 0,
+    fish: 0,
+    rabbit: 0
+  }
+
+  const fn = oneOfWeighted([
+    [0.8, 'cat'],
+    [0.025, 'dog'],
+    [0.1, 'bird'],
+    [0.025, 'fish'],
+    [0.05, 'rabbit']
+  ])
+
+  while (++i < n) sums[fn(i)]++
+
+  t.assert(diffBetween(sums.cat / n, 0.8) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.dog / n, 0.025) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.bird / n, 0.1) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.fish / n, 0.025) <= DIFF_THRESHOLD + 0.07)
+  t.assert(diffBetween(sums.rabbit / n, 0.05) <= DIFF_THRESHOLD)
+})
+
+test(`5/5 averages to within ${DIFF_THRESHOLD * 100}% of the given probabilities`, t => {
   const n = 10000
   let i = -1
 
   const sums = {
-    red: 0,
-    green: 0,
-    blue: 0
+    alpha: 0,
+    beta: 0,
+    gamma: 0,
+    delta: 0,
+    epsilon: 0
   }
 
   const fn = oneOfWeighted([
-    [0.6, 'red'],
-    [0.1, 'green'],
-    [0.3, 'blue']
+    [0.25, 'alpha'],
+    [0.25, 'beta'],
+    [0.25, 'gamma'],
+    [0.15, 'delta'],
+    [0.1, 'epsilon']
   ])
 
   while (++i < n) sums[fn(i)]++
 
-  t.assert(diffBetween(sums.red / n, 0.6) <= DIFF_THRESHOLD)
-  t.assert(diffBetween(sums.green / n, 0.1) <= DIFF_THRESHOLD)
-  t.assert(diffBetween(sums.blue / n, 0.3) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.alpha / n, 0.25) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.beta / n, 0.25) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.gamma / n, 0.25) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.delta / n, 0.15) <= DIFF_THRESHOLD)
+  t.assert(diffBetween(sums.epsilon / n, 0.1) <= DIFF_THRESHOLD)
 })
 
 test(`unassigned probabilities`, t => {