-
Notifications
You must be signed in to change notification settings - Fork 10
/
distance.go
191 lines (163 loc) · 5.17 KB
/
distance.go
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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
package som
import (
"container/heap"
"fmt"
"math"
"gonum.org/v1/gonum/mat"
)
// Metric is distance metric
type Metric int
const (
// Euclidean metric
Euclidean Metric = iota
)
// Distance calculates given metric distance between vectors a and b and returns it.
// If unsupported metric is requested it returns default distance which is Euclidean distance.
// It returns error if the supplied vectors are either nil or have different dimensions
func Distance(m Metric, a, b []float64) (float64, error) {
if a == nil || b == nil {
return 0.0, fmt.Errorf("invalid vectors supplied. a: %v, b: %v", a, b)
}
if len(a) != len(b) {
return 0.0, fmt.Errorf("incorrect vector dims. a: %d, b: %d", len(a), len(b))
}
switch m {
case Euclidean:
return euclideanVec(a, b), nil
default:
return euclideanVec(a, b), nil
}
}
// DistanceMx calculates given metric distance matrix for the supplied matrix.
// Distance matrix is also known in literature as dissimilarity matrix.
// DistanceMx returns a hollow symmetric matrix where an item x_ij stores the distance between
// vectors stored in rows i and j. If an unknown metric is supplied Euclidean distance is computed.
// It returns error if the supplied matrix is nil.
func DistanceMx(m Metric, mat *mat.Dense) (*mat.Dense, error) {
if mat == nil {
return nil, fmt.Errorf("invalid matrix supplied: %v", mat)
}
switch m {
case Euclidean:
return euclideanMx(mat), nil
default:
return euclideanMx(mat), nil
}
}
// ClosestVec finds the index of the closest vector to v in the list of vectors
// stored as rows in matrix m using the supplied distance metric.
// If unsupported metric is requested, ClosestVec falls over to euclidean metric.
// If several vectors of the same distance are found, it returns the index of the first one from the top.
// ClosestVec returns error if either v or m are nil or if the v dimension is different from
// the number of m columns. When the ClosestVec fails with error returned index is set to -1.
func ClosestVec(m Metric, v []float64, mat *mat.Dense) (int, error) {
if len(v) == 0 {
return -1, fmt.Errorf("invalid vector: %v", v)
}
if mat == nil {
return -1, fmt.Errorf("invalid matrix: %v", mat)
}
rows, _ := mat.Dims()
closest := 0
dist := math.MaxFloat64
for i := 0; i < rows; i++ {
d, err := Distance(m, v, mat.RawRowView(i))
if err != nil {
return -1, err
}
if d < dist {
dist = d
closest = i
}
}
return closest, nil
}
// ClosestNVec finds the N closest vectors to v in the list of vectors stored in m rows
// using the supplied distance metric. It returns a slice which contains indices to the m
// rows. The length of the slice is the same as number of requested closest vectors - n.
// ClosestNVec fails in the same way as ClosestVec. If n is higher than the number of
// rows in m, or if it is not a positive integer, it fails with error too.
func ClosestNVec(m Metric, n int, v []float64, mat *mat.Dense) ([]int, error) {
if len(v) == 0 {
return nil, fmt.Errorf("invalid vector: %v", v)
}
if mat == nil {
return nil, fmt.Errorf("invalid matrix: %v", mat)
}
rows, _ := mat.Dims()
if n <= 0 || n > rows {
return nil, fmt.Errorf("invalid number of closest vectors requested: %d", n)
}
closest := make([]int, n)
switch {
case n == 1:
idx, err := ClosestVec(m, v, mat)
if err != nil {
return nil, err
}
closest[0] = idx
default:
h, _ := newFloat64Heap(n)
rows, _ := mat.Dims()
for i := 0; i < rows; i++ {
d, err := Distance(m, v, mat.RawRowView(i))
if err != nil {
return nil, err
}
f := &float64Item{val: d, index: i}
heap.Push(h, f)
}
for j := 0; j < n; j++ {
closest[j] = (heap.Pop(h).(*float64Item)).index
}
}
return closest, nil
}
// BMUs returns a slice which contains indices of the Best Match Unit (BMU) codebook vectors for each
// vector stored in data rows. Each item in the returned slice correspnds to index of BMU in codebook for
// a particular data sample. If some data row has more than one BMU the index of the first one found is used.
// It returns error if either the data or codebook are nil or if their dimensions are mismatched.
func BMUs(data, codebook *mat.Dense) ([]int, error) {
if data == nil {
return nil, fmt.Errorf("invalid data supplied: %v", data)
}
if codebook == nil {
return nil, fmt.Errorf("invalid codebook supplied: %v", codebook)
}
rows, _ := data.Dims()
bmus := make([]int, rows)
for i := 0; i < rows; i++ {
idx, err := ClosestVec(Euclidean, data.RawRowView(i), codebook)
if err != nil {
return nil, err
}
bmus[i] = idx
}
return bmus, nil
}
// euclideanVec computes euclidean distance between vectors a and b.
func euclideanVec(a, b []float64) float64 {
d := 0.0
for i := 0; i < len(a); i++ {
d += (a[i] - b[i]) * (a[i] - b[i])
}
return math.Sqrt(d)
}
// euclideanMx computes a matrix of euclidean distances between each row in m
func euclideanMx(m *mat.Dense) *mat.Dense {
rows, _ := m.Dims()
out := mat.NewDense(rows, rows, nil)
for row := 0; row < rows-1; row++ {
dist := 0.0
a := m.RawRowView(row)
for i := row + 1; i < rows; i++ {
if i != row {
b := m.RawRowView(i)
dist = euclideanVec(a, b)
out.Set(row, i, dist)
out.Set(i, row, dist)
}
}
}
return out
}