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
|
package native
import (
"fmt"
. "gorgonia.org/tensor"
)
// There are times where it is more effective to use native Go slice semantics to do work (for example, when performing batch work over kernels).
// Iterators are useful for this purpose. This package provides iterators for the standard types
// However, custom types are also available. See Vector, Matrix and Tensor3 examples.
func Example_iterator() {
var T *Dense
T = New(WithShape(2, 3), WithBacking(Range(Float64, 0, 6)))
x, err := MatrixF64(T)
if err != nil {
fmt.Printf("ERR: %v", err)
}
for _, row := range x {
fmt.Printf("%v\n", row)
}
// Output:
// [0 1 2]
// [3 4 5]
}
// The NativeSelect function squashes the dimensions, and returns an iterator in native Go slice semantics.
func Example_select() {
// Selection is a bit of an interesting use case. Sometimes you don't want to iterate through the layers.
//
// For example, in a number of use cases where you have a 4-Tensor, you'd typically reshape it to some
// 2D matrix which can then be plugged into BLAS algorithms directly. Sometimes you wouldn't need to reshape.
// All you have to do is squash the dimensions inwards. This function does that.
//
// The best way to explain the Select functions is through concrete examples.
// Imagine a tensor with (2,3,4,5) shape. Arbitrarily, we call them (NCHW) - Batch Size, Channel Count, Height, Width.
// If we want to select all the channels, across all batches, then `NativeSelectX(T, 1)` would yield all channels. The resulting matrix will be (6, 20)
// If we want to select all the heights, across all channels and batches, then `NativeSelectX(T, 2) will yield all heights. The resulting matrix will be (24, 5)
//
// If for some reason the format was in NHWC, then you would need to reshape. This wouldn't be useful.
var T *Dense
T = New(WithShape(2, 3, 4, 5), WithBacking(Range(Float64, 0, 2*3*4*5)))
x, err := SelectF64(T, 1)
if err != nil {
fmt.Printf("ERR %v", err)
}
for _, row := range x {
fmt.Printf("%3.0f\n", row)
}
// Output:
// [ 0 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]
}
// The iterators are iteratos in the truest sense. The data isn't copied, as this example shows
func Example_clobber() {
var T *Dense
T = New(WithShape(2, 3), WithBacking(Range(Float64, 0, 6)))
fmt.Printf("Before :\n%v", T)
xx, _ := MatrixF64(T)
xx[1][1] = 10000
fmt.Printf("After :\n%v", T)
// Output:
// Before :
// ⎡0 1 2⎤
// ⎣3 4 5⎦
// After :
// ⎡ 0 1 2⎤
// ⎣ 3 10000 5⎦
}
|
