/
cure.js
177 lines (167 loc) · 4.23 KB
/
cure.js
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
import Matrix from '../util/matrix.js'
/**
* @typedef {object} CURENode
* @property {number[]} [point] Data point of leaf node
* @property {number} [index] Data index of leaf node
* @property {number[][]} [repr] Represented points
* @property {number} [distance] Distance between children nodes
* @property {number} size Number of leaf nodes
* @property {CURENode[]} [children] Children nodes
* @property {CURENode[]} leafs Leaf nodes
*/
/**
* Clustering Using REpresentatives
*/
export default class CURE {
// https://en.wikipedia.org/wiki/CURE_algorithm
// http://ibisforest.org/index.php?CURE
/**
* @param {number} c Number of representative points
*/
constructor(c) {
this._c = c
this._a = 0.2
this._root = null
}
_distance(a, b) {
return Math.sqrt(a.reduce((s, v, i) => s + (v - b[i]) ** 2, 0))
}
/**
* Fit model.
*
* @param {Array<Array<number>>} data Training data
*/
fit(data) {
const clusters = []
const distances = []
data.forEach((v, i) => {
clusters.push({
point: v,
index: i,
repr: [v],
distance: 0,
get leafs() {
return [this]
},
})
distances[i] = data.map(p => this._distance(v, p))
})
while (clusters.length > 1) {
let min_i = 0
let min_j = 1
let min_v = Infinity
const n = clusters.length
for (let i = 0; i < n; i++) {
for (let j = i + 1; j < n; j++) {
if (distances[i][j] < min_v) {
min_i = i
min_j = j
min_v = distances[i][j]
}
}
}
const i_datas = clusters[min_i].leafs.map(v => v.point)
const j_datas = clusters[min_j].leafs.map(v => v.point)
const new_datas = [...i_datas, ...j_datas]
const repr_idx = []
let pre_i = Math.floor(Math.random() * new_datas.length)
for (let i = 0; i < Math.min(new_datas.length, this._c); i++) {
let max_d = 0
let max_i = -1
for (let k = 0; k < new_datas.length; k++) {
if (repr_idx.includes(k)) {
continue
}
const d = this._distance(new_datas[k], new_datas[pre_i])
if (d > max_d) {
max_d = d
max_i = k
}
}
repr_idx.push(max_i)
pre_i = max_i
}
const repr = []
const mean = Matrix.fromArray(new_datas).mean(0).value
for (let i = 0; i < repr_idx.length; i++) {
repr[i] = new_datas[repr_idx[i]].concat()
for (let d = 0; d < mean.length; d++) {
repr[i][d] = this._a * mean[d] + (1 - this._a) * repr[i][d]
}
}
for (let i = 0; i < n; i++) {
if (i === min_i || i === min_j) {
distances[min_i][i] = 0
continue
}
let md = Infinity
const iv = clusters[i]
for (let s = 0; s < iv.repr.length; s++) {
for (let t = 0; t < repr.length; t++) {
const d = this._distance(iv.repr[s], repr[t])
if (d < md) {
md = d
}
}
}
distances[i][min_i] = distances[min_i][i] = md
distances[i].splice(min_j, 1)
}
distances[min_i].splice(min_j, 1)
distances.splice(min_j, 1)
clusters[min_i] = {
repr: repr,
distance: min_v,
children: [clusters[min_i], clusters[min_j]],
get leafs() {
return [...this.children[0].leafs, ...this.children[1].leafs]
},
}
clusters.splice(min_j, 1)
}
this._root = clusters[0]
}
/**
* Returns the specified number of clusters.
*
* @param {number} number Number of clusters
* @returns {CURENode[]} Cluster nodes
*/
getClusters(number) {
const scanNodes = [this._root]
while (scanNodes.length < number) {
let max_distance = 0
let max_distance_idx = -1
for (let i = 0; i < scanNodes.length; i++) {
const node = scanNodes[i]
if (node.children && node.distance > max_distance) {
max_distance_idx = i
max_distance = node.distance
}
}
if (max_distance_idx === -1) {
break
}
const max_distance_node = scanNodes[max_distance_idx]
scanNodes.splice(max_distance_idx, 1, max_distance_node.children[0], max_distance_node.children[1])
}
return scanNodes
}
/**
* Returns predicted categories.
*
* @param {number} k Number of clusters
* @returns {number[]} Predicted values
*/
predict(k) {
const p = []
const clusters = this.getClusters(k)
for (let i = 0; i < clusters.length; i++) {
const leafs = clusters[i].leafs
for (let k = 0; k < leafs.length; k++) {
p[leafs[k].index] = i
}
}
return p
}
}