-
Notifications
You must be signed in to change notification settings - Fork 5
/
multinom.go
250 lines (218 loc) · 5.11 KB
/
multinom.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
package spn
import (
"bytes"
"fmt"
"github.com/RenatoGeh/gospn/sys"
"math"
)
// Mode of a univariate distribution.
type Mode struct {
// Value of variable when it is the highest.
index int
// Highest value of variable.
val float64
}
func computeMode(pr []float64) (int, float64) {
var i int
var m float64
for j, p := range pr {
if p > m {
i, m = j, p
}
}
return i, m
}
// Multinomial represents a multinomial distribution.
type Multinomial struct {
Node
// Variable ID
varid int
// Discrete probability distribution
pr []float64
// Mode of pr. We pre-compute this to save time.
mode Mode
}
// NewMultinomial constructs a new Multinomial.
func NewMultinomial(varid int, dist []float64) *Multinomial {
n := len(dist)
m := -1.0
miv := make([]int, n)
var mi int
nmi := 0
for i := 0; i < n; i++ {
if dist[i] > m {
m = dist[i]
miv[0] = i
mi = i
nmi = 1
} else {
if dist[i] == m {
miv[nmi] = i
nmi++
}
}
}
if nmi > 1 {
mi = miv[sys.RandIntn(nmi)]
}
return &Multinomial{Node{sc: []int{varid}}, varid, dist, Mode{mi, m}}
}
// NewCountingMultinomial constructs a new Multinomial from a count slice.
func NewCountingMultinomial(varid int, counts []int) *Multinomial {
n := len(counts)
pr := make([]float64, n)
s := 0.0
for i := 0; i < n; i++ {
s += 1.0 + float64(counts[i])
pr[i] = float64(1 + counts[i])
}
for i := 0; i < n; i++ {
pr[i] /= float64(s)
}
m := -1.0
var mi int
miv := make([]int, n)
nmi := 0
for i := 0; i < n; i++ {
if pr[i] > m {
m = pr[i]
miv[0] = i
mi = i
nmi = 1
} else {
if pr[i] == m {
miv[nmi] = i
nmi++
}
}
}
if nmi > 1 {
mi = miv[sys.RandIntn(nmi)]
}
return &Multinomial{Node{sc: []int{varid}}, varid, pr, Mode{mi, m}}
}
// NewScopedCountingMultinomial does the same as NewCountingMultinomial except it allows multiple
// variable scope.
func NewScopedCountingMultinomial(varid int, esc []int, counts []int) *Multinomial {
n := len(counts)
pr := make([]float64, n)
s := 0.0
for i := 0; i < n; i++ {
s += 1.0 + float64(counts[i])
pr[i] = float64(1 + counts[i])
}
for i := 0; i < n; i++ {
pr[i] /= float64(s)
}
var m float64
var mi int
for i := 0; i < n; i++ {
if pr[i] > m {
m = pr[i]
mi = i
}
}
return &Multinomial{Node{sc: esc}, varid, pr, Mode{mi, m}}
}
// NewEmptyMultinomial constructs a new empty Multinomial for learning. Argument m is the
// cardinality of varid.
func NewEmptyMultinomial(varid, m int) *Multinomial {
pr := make([]float64, m)
for i := 0; i < m; i++ {
pr[i] = 1.0 / float64(m)
}
return &Multinomial{Node{sc: []int{varid}}, varid, pr, Mode{0, pr[0]}}
}
// Type returns the type of this node.
func (m *Multinomial) Type() string { return "leaf" }
// Pr returns the discrete probability distribution.
func (m *Multinomial) Pr() []float64 { return m.pr }
// Value returns the probability of a certain valuation. That is Pr(X=val[varid]), where
// Pr is a probability function over a Multinomial distribution.
func (m *Multinomial) Value(val VarSet) float64 {
v, ok := val[m.varid]
if ok {
return math.Log(m.pr[v])
}
return 0.0 // ln(1.0) = 0.0
}
// Max returns the MAP state given a valuation.
func (m *Multinomial) Max(val VarSet) float64 {
v, ok := val[m.varid]
if ok {
return math.Log(m.pr[v])
}
return math.Log(m.mode.val)
}
// ArgMax returns both the arguments and the value of the MAP state given a certain valuation.
func (m *Multinomial) ArgMax(val VarSet) (VarSet, float64) {
retval := make(VarSet)
v, ok := val[m.varid]
if ok {
retval[m.varid] = v
return retval, math.Log(m.pr[v])
}
retval[m.varid] = m.mode.index
return retval, math.Log(m.mode.val)
}
// Mean returns the mean of the distribution.
func (m *Multinomial) Mean() float64 {
var mu float64
for i, p := range m.pr {
mu += p * float64(i)
}
return mu
}
// StdDev returns the standard deviation of the distribution.
func (m *Multinomial) StdDev() float64 {
mu := m.Mean()
var sd float64
for i, _ := range m.pr {
d := float64(i) - mu
sd += d * d
}
sd = math.Sqrt(sd / float64(len(m.pr)))
return sd
}
// MuSigma returns the mean and standard deviation of the distribution.
func (m *Multinomial) MuSigma() (float64, float64) {
mu := m.Mean()
var sd float64
for i, _ := range m.pr {
d := float64(i) - mu
sd += d * d
}
sd = math.Sqrt(sd / float64(len(m.pr)))
return mu, sd
}
// GobEncode serializes this multinomial node.
func (m *Multinomial) GobEncode() ([]byte, error) {
var b bytes.Buffer
fmt.Fprintf(&b, "%d %d", m.varid, len(m.pr))
for _, p := range m.pr {
fmt.Fprintf(&b, " %f", p)
}
return b.Bytes(), nil
}
// GobDecode unserializes this multinomial node.
func (m *Multinomial) GobDecode(data []byte) error {
b := bytes.NewBuffer(data)
var n int
_, err := fmt.Fscanf(b, "%d %d", &m.varid, &n)
if err != nil {
return err
}
m.sc = []int{m.varid}
m.pr = make([]float64, n)
for i := 0; i < n; i++ {
_, err = fmt.Fscanf(b, "%f", &m.pr[i])
if err != nil {
return err
}
}
imode, mode := computeMode(m.pr)
m.mode.index, m.mode.val = imode, mode
return nil
}
// SubType returns this leaf's subtype.
func (m *Multinomial) SubType() string { return "multinomial" }