Scatter - 9 vs 11#
Next section compares an older to a newer version of the same operator after both definition are converted into markdown text. Green means an addition to the newer version, red means a deletion. Anything else is unchanged.
- Scatter9 → Scatter11 +31 -57
Scatter9 → Scatter11
RENAMED
@@ -1 +1 @@
|
|
1
|
+
Given data, updates and indices input tensors of rank r >= 1, write the values provided by updates
|
2
|
+
into the first input, data, along axis dimension of data (by default outer-most one as axis=0) at corresponding indices.
|
1
|
-
This operator is deprecated. Please use ScatterElements, which provides the same functionality.
|
2
|
-
|
3
|
-
Scatter takes three inputs data, updates, and indices of the same
|
4
|
-
rank r >= 1 and an optional attribute axis that identifies an axis of data
|
5
|
-
(by default, the outer-most axis, that is axis 0). The output of the operation
|
6
|
-
is produced by creating a copy of the input data, and then updating its value
|
7
|
-
to values specified by updates at specific index positions specified by
|
8
|
-
indices. Its output shape is the same as the shape of data.
|
9
|
-
|
10
|
-
For each entry in updates, the target index in data is
|
3
|
+
For each entry in updates, the target index in data is specified by corresponding entry in indices
|
4
|
+
for dimension = axis, and index in source for dimension != axis. For instance, in a 2-D tensor case,
|
5
|
+
data[indices[i][j]][j] = updates[i][j] if axis = 0, or data[i][indices[i][j]] = updates[i][j] if axis = 1,
|
6
|
+
where i and j are loop counters from 0 up to the respective size in updates - 1.
|
11
|
-
the corresponding entry in indices with the index of the entry itself: the
|
12
|
-
index-value for dimension = axis is obtained from the value of the corresponding
|
13
|
-
entry in indices and the index-value for dimension != axis is obtained from the
|
14
|
-
index of the entry itself.
|
15
|
-
|
16
|
-
For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry
|
17
|
-
is performed as below:
|
18
|
-
::
|
19
|
-
|
20
|
-
output[indices[i][j]][j] = updates[i][j] if axis = 0,
|
21
|
-
output[i][indices[i][j]] = updates[i][j] if axis = 1,
|
22
|
-
|
23
|
-
This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation.
|
24
|
-
|
25
7
|
Example 1:
|
26
|
-
::
|
27
|
-
|
28
|
-
|
8
|
+
data = [
|
29
|
-
|
9
|
+
[0.0, 0.0, 0.0],
|
30
|
-
|
10
|
+
[0.0, 0.0, 0.0],
|
31
|
-
|
11
|
+
[0.0, 0.0, 0.0],
|
32
|
-
|
12
|
+
]
|
33
|
-
|
13
|
+
indices = [
|
34
|
-
|
14
|
+
[1, 0, 2],
|
35
|
-
|
15
|
+
[0, 2, 1],
|
36
|
-
|
16
|
+
]
|
37
|
-
|
17
|
+
updates = [
|
38
|
-
|
18
|
+
[1.0, 1.1, 1.2],
|
39
|
-
|
19
|
+
[2.0, 2.1, 2.2],
|
40
|
-
|
20
|
+
]
|
41
|
-
|
21
|
+
output = [
|
42
|
-
|
22
|
+
[2.0, 1.1, 0.0]
|
43
|
-
|
23
|
+
[1.0, 0.0, 2.2]
|
44
|
-
|
24
|
+
[0.0, 2.1, 1.2]
|
25
|
+
]
|
45
|
-
]
|
46
|
-
|
47
26
|
Example 2:
|
48
|
-
::
|
49
|
-
|
50
|
-
|
27
|
+
data = [[1.0, 2.0, 3.0, 4.0, 5.0]]
|
51
|
-
|
28
|
+
indices = [[1, 3]]
|
52
|
-
|
29
|
+
updates = [[1.1, 2.1]]
|
53
|
-
|
30
|
+
axis = 1
|
54
|
-
|
31
|
+
output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
|
55
32
|
**Attributes**
|
56
33
|
* **axis**:
|
57
34
|
Which axis to scatter on. Negative value means counting dimensions
|
58
|
-
from the back. Accepted range is [-r, r-1]
|
35
|
+
from the back. Accepted range is [-r, r-1]
|
59
36
|
**Inputs**
|
60
37
|
* **data** (heterogeneous) - **T**:
|
61
38
|
Tensor of rank r >= 1.
|
62
39
|
* **indices** (heterogeneous) - **Tind**:
|
63
|
-
Tensor of int32/int64 indices, of r >= 1 (same rank as input).
|
40
|
+
Tensor of int32/int64 indices, of r >= 1 (same rank as input).
|
64
|
-
index values are expected to be within bounds [-s, s-1] along axis
|
65
|
-
of size s. It is an error if any of the index values are out of
|
66
|
-
bounds.
|
67
41
|
* **updates** (heterogeneous) - **T**:
|
68
42
|
Tensor of rank r >=1 (same rank and shape as indices)
|
69
43
|
**Outputs**
|
70
44
|
* **output** (heterogeneous) - **T**:
|
71
45
|
Tensor of rank r >= 1 (same rank as input).
|
72
46
|
**Type Constraints**
|
73
47
|
* **T** in (
|
74
48
|
tensor(bool),
|
75
49
|
tensor(complex128),
|
76
50
|
tensor(complex64),
|
77
51
|
tensor(double),
|
78
52
|
tensor(float),
|
79
53
|
tensor(float16),
|
80
54
|
tensor(int16),
|
81
55
|
tensor(int32),
|
82
56
|
tensor(int64),
|
83
57
|
tensor(int8),
|
84
58
|
tensor(string),
|
85
59
|
tensor(uint16),
|
86
60
|
tensor(uint32),
|
87
61
|
tensor(uint64),
|
88
62
|
tensor(uint8)
|
89
63
|
):
|
90
64
|
Input and output types can be of any tensor type.
|
91
65
|
* **Tind** in (
|
92
66
|
tensor(int32),
|
93
67
|
tensor(int64)
|
94
68
|
):
|
95
69
|
Constrain indices to integer types
|