-
Notifications
You must be signed in to change notification settings - Fork 1
/
confusion.go
281 lines (236 loc) · 6.17 KB
/
confusion.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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
package mlmetrics
import (
"math"
"sync"
)
// ConfusionMatrix can be used to visualize the performance of a binary
// classifier.
type ConfusionMatrix struct {
mat resizableMatrix
mu sync.RWMutex
}
// NewConfusionMatrix inits a new ConfusionMatrix.
func NewConfusionMatrix() *ConfusionMatrix {
return new(ConfusionMatrix)
}
// Reset resets the state.
func (m *ConfusionMatrix) Reset() {
m.mu.Lock()
m.mat = resizableMatrix{}
m.mu.Unlock()
}
// Observe records an observation of the actual vs the predicted category.
func (m *ConfusionMatrix) Observe(actual, predicted int) {
m.ObserveWeight(actual, predicted, 1.0)
}
// ObserveWeight records an observation of the actual vs the predicted category with a given weight.
func (m *ConfusionMatrix) ObserveWeight(actual, predicted int, weight float64) {
if !isValidCategory(actual) || !isValidCategory(predicted) || !isValidWeight(weight) {
return
}
m.mu.Lock()
m.mat.Set(actual, predicted, m.mat.At(actual, predicted)+weight)
m.mu.Unlock()
}
// Order returns the matrix order (number or rows/cols).
func (m *ConfusionMatrix) Order() int {
m.mu.RLock()
size := m.mat.size
m.mu.RUnlock()
return size
}
// TotalWeight returns the total weight observed (sum of the matrix).
func (m *ConfusionMatrix) TotalWeight() float64 {
m.mu.RLock()
sum := m.mat.Sum()
m.mu.RUnlock()
return sum
}
// Row returns the distribution of predicted weights for category x.
func (m *ConfusionMatrix) Row(x int) []float64 {
m.mu.RLock()
defer m.mu.RUnlock()
if x >= m.mat.size {
return nil
}
row := make([]float64, m.mat.size)
copy(row, m.mat.Row(x))
return row
}
// Column returns the distribution of actual weights for category x.
func (m *ConfusionMatrix) Column(x int) []float64 {
m.mu.RLock()
defer m.mu.RUnlock()
if x >= m.mat.size {
return nil
}
col := make([]float64, m.mat.size)
for i := 0; i < m.mat.size; i++ {
col[i] = m.mat.At(i, x)
}
return col
}
// Accuracy returns the overall accuracy rate.
func (m *ConfusionMatrix) Accuracy() float64 {
m.mu.RLock()
defer m.mu.RUnlock()
sum := m.mat.Sum()
if sum == 0.0 {
return 0.0
}
var pos float64
for i := 0; i < m.mat.size; i++ {
pos += m.mat.At(i, i)
}
return pos / sum
}
// Precision calculates the positive predictive value for category x.
func (m *ConfusionMatrix) Precision(x int) float64 {
m.mu.RLock()
defer m.mu.RUnlock()
sum := m.mat.ColSum(x)
if sum == 0.0 {
return 0.0
}
return m.mat.At(x, x) / sum
}
// Sensitivity calculates the recall (aka 'hit rate') for category x.
func (m *ConfusionMatrix) Sensitivity(x int) float64 {
m.mu.RLock()
defer m.mu.RUnlock()
sum := m.mat.RowSum(x)
if sum == 0.0 {
return 0.0
}
return m.mat.At(x, x) / sum
}
// F1 calculates the F1 score for category x, the harmonic mean of precision and sensitivity.
func (m *ConfusionMatrix) F1(x int) float64 {
m.mu.RLock()
defer m.mu.RUnlock()
csm := m.mat.ColSum(x)
if csm == 0 {
return 0
}
rsm := m.mat.RowSum(x)
if rsm == 0 {
return 0
}
pos := m.mat.At(x, x)
precision := pos / csm
sensitivity := pos / rsm
return 2 * precision * sensitivity / (precision + sensitivity)
}
// Kappa represents the Cohen's Kappa, a statistic which measures inter-rater agreement for qualitative
// (categorical) items. It is generally thought to be a more robust measure than simple percent agreement
// calculation, as κ takes into account the possibility of the agreement occurring by chance.
// https://en.wikipedia.org/wiki/Cohen%27s_kappa
func (m *ConfusionMatrix) Kappa() float64 {
m.mu.RLock()
defer m.mu.RUnlock()
sum := m.mat.Sum()
if sum == 0.0 {
return 0.0
}
var obs, exp float64
for i := 0; i < m.mat.size; i++ {
obs += m.mat.At(i, i)
exp += m.mat.RowSum(i) * m.mat.ColSum(i) / sum
}
if div := sum - exp; div != 0 {
return (obs - exp) / div
}
return 1.0
}
// Matthews is a correlation coefficient used as a measure of the quality of binary
// and multiclass classifications. It takes into account true and false positives
// and negatives and is generally regarded as a balanced measure which can be
// used even if the classes are of very different sizes. The MCC is in essence
// a correlation coefficient value between -1 and +1. A coefficient of +1 represents
// a perfect prediction, 0 an average random prediction and -1 an inverse prediction.
// The statistic is also known as the phi coefficient. [source: Wikipedia]
func (m *ConfusionMatrix) Matthews() float64 {
m.mu.RLock()
defer m.mu.RUnlock()
sum := m.mat.Sum()
if sum == 0.0 {
return 0.0
}
var exp, cf1, cf2, cf3 float64
for i := 0; i < m.mat.size; i++ {
rsum := m.mat.RowSum(i)
csum := m.mat.ColSum(i)
exp += m.mat.At(i, i)
cf1 += rsum * csum
cf2 += rsum * rsum
cf3 += csum * csum
}
sum2 := sum * sum
if pdt := (sum2 - cf2) * (sum2 - cf3); pdt != 0 {
return ((exp * sum) - cf1) / math.Sqrt(pdt)
}
return 0
}
type resizableMatrix struct {
size int
data []float64
}
// Set sets the field at (i, j) to v
func (m *resizableMatrix) Set(i, j int, v float64) {
m.resize(maxInt(i+1, j+1))
m.data[i*m.size+j] = v
}
// At returns the value at (i, j)
func (m *resizableMatrix) At(i, j int) float64 {
if i < m.size && j < m.size {
return m.data[i*m.size+j]
}
return 0
}
// Row returns the slice of row at i.
func (m *resizableMatrix) Row(i int) []float64 {
if i >= 0 && i < m.size {
offset := i * m.size
return m.data[offset : offset+m.size]
}
return nil
}
// RowSum returns the sum of values in row (i).
func (m *resizableMatrix) RowSum(i int) (sum float64) {
if i >= 0 && i < m.size {
offset := (i * m.size)
for k := offset; k < offset+m.size; k++ {
sum += m.data[k]
}
}
return
}
// RowSum returns the sum of values in col (j).
func (m *resizableMatrix) ColSum(j int) (sum float64) {
if j >= 0 && j < m.size {
for k := j; k < len(m.data); k += m.size {
sum += m.data[k]
}
}
return
}
// Sum calculates the sum of all cells.
func (m *resizableMatrix) Sum() float64 {
sum := 0.0
for _, v := range m.data {
sum += v
}
return sum
}
func (m *resizableMatrix) resize(n int) {
if n <= m.size {
return
}
data := make([]float64, n*n)
for row := 0; row < m.size; row++ {
offset := row * m.size
copy(data[row*n:], m.data[offset:offset+m.size])
}
m.size = n
m.data = data
}