ScatterElements - 13 vs 16

Files changed (1) hide show
  1. ScatterElements13 → ScatterElements16 +21 -5
ScatterElements13 → ScatterElements16 RENAMED
@@ -1 +1 @@
1
1
  ScatterElements takes three inputs data, updates, and indices of the same
2
2
  rank r >= 1 and an optional attribute axis that identifies an axis of data
3
3
  (by default, the outer-most axis, that is axis 0). The output of the operation
4
4
  is produced by creating a copy of the input data, and then updating its value
5
5
  to values specified by updates at specific index positions specified by
6
6
  indices. Its output shape is the same as the shape of data.
7
-
8
7
  For each entry in updates, the target index in data is obtained by combining
9
8
  the corresponding entry in indices with the index of the entry itself: the
10
9
  index-value for dimension = axis is obtained from the value of the corresponding
11
10
  entry in indices and the index-value for dimension != axis is obtained from the
12
11
  index of the entry itself.
13
-
12
+ reduction allows specification of an optional reduction operation, which is applied to all values in updates
13
+ tensor into output at the specified indices.
14
+ In cases where reduction is set to "none", indices should not have duplicate entries: that is, if idx1 != idx2,
14
- For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry
15
+ then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update
15
- is performed as below:
16
+ corresponding to the [i][j] entry is performed as below:
16
17
  ::
17
18
  output[indices[i][j]][j] = updates[i][j] if axis = 0,
18
19
  output[i][indices[i][j]] = updates[i][j] if axis = 1,
20
+ When reduction is set to "add", the update corresponding to the [i][j] entry is performed as below:
21
+ ::
22
+
23
+ output[indices[i][j]][j] += updates[i][j] if axis = 0,
24
+ output[i][indices[i][j]] += updates[i][j] if axis = 1,
25
+
26
+ When reduction is set to "mul", the update corresponding to the [i][j] entry is performed as below:
27
+ ::
28
+
29
+ output[indices[i][j]][j] *= updates[i][j] if axis = 0,
30
+ output[i][indices[i][j]] *= updates[i][j] if axis = 1,
31
+
19
32
  This operator is the inverse of GatherElements. It is similar to Torch's Scatter operation.
20
-
21
33
  Example 1:
22
34
  ::
23
35
  data = [
24
36
  [0.0, 0.0, 0.0],
25
37
  [0.0, 0.0, 0.0],
26
38
  [0.0, 0.0, 0.0],
27
39
  ]
28
40
  indices = [
29
41
  [1, 0, 2],
30
42
  [0, 2, 1],
31
43
  ]
32
44
  updates = [
33
45
  [1.0, 1.1, 1.2],
34
46
  [2.0, 2.1, 2.2],
35
47
  ]
36
48
  output = [
37
49
  [2.0, 1.1, 0.0]
38
50
  [1.0, 0.0, 2.2]
39
51
  [0.0, 2.1, 1.2]
40
52
  ]
41
53
  Example 2:
42
54
  ::
43
55
  data = [[1.0, 2.0, 3.0, 4.0, 5.0]]
44
56
  indices = [[1, 3]]
45
57
  updates = [[1.1, 2.1]]
46
58
  axis = 1
47
59
  output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
48
60
  **Attributes**
49
61
  * **axis**:
50
62
  Which axis to scatter on. Negative value means counting dimensions
51
63
  from the back. Accepted range is [-r, r-1] where r = rank(data).
64
+ * **reduction**:
65
+ Type of reduction to apply: none (default), add, mul. 'none': no
66
+ reduction applied. 'add': reduction using the addition operation.
67
+ 'mul': reduction using the multiplication operation.
52
68
  **Inputs**
53
69
  * **data** (heterogeneous) - **T**:
54
70
  Tensor of rank r >= 1.
55
71
  * **indices** (heterogeneous) - **Tind**:
56
72
  Tensor of int32/int64 indices, of r >= 1 (same rank as input). All
57
73
  index values are expected to be within bounds [-s, s-1] along axis
58
74
  of size s. It is an error if any of the index values are out of
59
75
  bounds.
60
76
  * **updates** (heterogeneous) - **T**:
61
77
  Tensor of rank r >=1 (same rank and shape as indices)
62
78
  **Outputs**
63
79
  * **output** (heterogeneous) - **T**:
64
80
  Tensor of rank r >= 1 (same rank as input).
65
81
  **Type Constraints**
66
82
  * **T** in (
67
83
  tensor(bfloat16),
68
84
  tensor(bool),
69
85
  tensor(complex128),
70
86
  tensor(complex64),
71
87
  tensor(double),
72
88
  tensor(float),
73
89
  tensor(float16),
74
90
  tensor(int16),
75
91
  tensor(int32),
76
92
  tensor(int64),
77
93
  tensor(int8),
78
94
  tensor(string),
79
95
  tensor(uint16),
80
96
  tensor(uint32),
81
97
  tensor(uint64),
82
98
  tensor(uint8)
83
99
  ):
84
100
  Input and output types can be of any tensor type.
85
101
  * **Tind** in (
86
102
  tensor(int32),
87
103
  tensor(int64)
88
104
  ):
89
105
  Constrain indices to integer types