forked from pa-m/sklearn
/
split.go
154 lines (135 loc) · 3.64 KB
/
split.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
package modelselection
import (
"math"
"golang.org/x/exp/rand"
"gonum.org/v1/gonum/mat"
"github.com/etrace-io/sklearn/base"
)
// RandomState is to init a new random source for reproducibility
type RandomState = rand.Rand
// KFold ...
type KFold struct {
NSplits int
Shuffle bool
RandomState base.RandomState
}
var (
_ Splitter = &KFold{}
)
// Splitter is the interface for splitters like KFold
type Splitter interface {
Split(X, Y *mat.Dense) (ch chan Split)
GetNSplits(X, Y *mat.Dense) int
SplitterClone() Splitter
}
// Split ...
type Split struct{ TrainIndex, TestIndex []int }
// SplitterClone ...
func (splitter *KFold) SplitterClone() Splitter {
if splitter == nil {
return nil
}
clone := *splitter
if sourceCloner, ok := clone.RandomState.(base.SourceCloner); ok && sourceCloner != base.SourceCloner(nil) {
clone.RandomState = sourceCloner.SourceClone()
}
return &clone
}
// Split generate Split structs
func (splitter *KFold) Split(X, Y *mat.Dense) (ch chan Split) {
if splitter.NSplits <= 0 {
splitter.NSplits = 3
}
NSamples, _ := X.Dims()
type Shuffler interface {
Shuffle(n int, swap func(i, j int))
}
type Intner interface{ Intn(int) int }
var rndShuffle = rand.Shuffle
var rndIntn = rand.Intn
if splitter.RandomState != base.Source(nil) {
if shuffler, ok := splitter.RandomState.(Shuffler); ok {
rndShuffle = shuffler.Shuffle
} else {
rndShuffle = rand.New(splitter.RandomState).Shuffle
}
if intner, ok := splitter.RandomState.(Intner); ok {
rndIntn = intner.Intn
} else {
rndIntn = rand.New(splitter.RandomState).Intn
}
}
ch = make(chan Split)
go func() {
for isplit := 0; isplit < splitter.NSplits; isplit++ {
NTest := NSamples / splitter.NSplits
// The first n_samples % n_splits folds have size n_samples // n_splits + 1, other folds have size n_samples // n_splits, where n_samples is the number of samples.
if isplit < NSamples%splitter.NSplits {
NTest++
}
a := make([]int, NSamples)
for i := range a {
a[i] = i
}
aSwap := func(i, j int) { a[i], a[j] = a[j], a[i] }
if splitter.Shuffle {
rndShuffle(len(a), aSwap)
} else {
start := rndIntn(NSamples)
for i := 0; i < NTest; i++ {
aSwap((start+i)%NSamples, NSamples-NTest+i)
}
}
sp := Split{
TrainIndex: a[:NSamples-NTest],
TestIndex: a[NSamples-NTest:],
}
ch <- sp
}
close(ch)
}()
return ch
}
// GetNSplits for KFold
func (splitter *KFold) GetNSplits(X, Y *mat.Dense) int {
return splitter.NSplits
}
// TrainTestSplit splits X and Y into test set and train set
// testsize must be between 0 and 1
// it produce same sets than scikit-learn
func TrainTestSplit(X, Y mat.Matrix, testsize float64, randomstate uint64) (Xtrain, Xtest, ytrain, ytest *mat.Dense) {
NSamples, NFeatures := X.Dims()
_, NOutputs := Y.Dims()
var testlen int
if testsize > 1 {
testlen = int(math.Ceil(math.Min(float64(NSamples), testsize)))
} else {
testlen = int(math.Ceil(float64(NSamples) * testsize))
}
Xtest = mat.NewDense(testlen, NFeatures, nil)
ytest = mat.NewDense(testlen, NOutputs, nil)
Xtrain = mat.NewDense(NSamples-testlen, NFeatures, nil)
ytrain = mat.NewDense(NSamples-testlen, NOutputs, nil)
src := base.NewLockedSource(randomstate)
var ind []int
src.WithLock(func(src base.Source) {
permer, ok := src.(base.Permer)
if !ok {
panic("Source does not implement Perm")
}
{
ind = permer.Perm(NSamples)
}
})
for i := 0; i < NSamples; i++ {
j := ind[i]
if i < testlen {
mat.Row(Xtest.RawRowView(i), j, X)
mat.Row(ytest.RawRowView(i), j, Y)
} else {
mat.Row(Xtrain.RawRowView(i-testlen), j, X)
mat.Row(ytrain.RawRowView(i-testlen), j, Y)
}
}
return
}