GatherND - 11 vs 12

Files changed (1) hide show
  1. GatherND11 → GatherND12 +44 -13
GatherND11 → GatherND12 RENAMED
@@ -1 +1 @@
1
- Given data tensor of rank r >= 1, and indices tensor of rank q >= 1, this operator gathers
1
+ Given data tensor of rank r >= 1, indices tensor of rank q >= 1, and batch_dims integer b, this operator gathers
2
- slices of data into an output tensor of rank q + r - indices_shape[-1] - 1.
2
+ slices of data into an output tensor of rank q + r - indices_shape[-1] - 1 - b.
3
3
  indices is an q-dimensional integer tensor, best thought of as a (q-1)-dimensional tensor of index-tuples into data,
4
4
  where each element defines a slice of data
5
+
6
+ batch_dims (denoted as b) is an integer indicating the number of batch dimensions, i.e the leading b number of dimensions of
7
+ data tensor and indices are representing the batches, and the gather starts from the b+1 dimension.
5
8
  Some salient points about the inputs' rank and shape:
6
9
  1) r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks r and q
10
+ 2) The first b dimensions of the shape of indices tensor and data tensor must be equal.
11
+ 3) b < min(q, r) is to be honored.
12
+
7
- 2) The indices_shape[-1] should have a value between 1 (inclusive) and rank r (inclusive)
13
+ 4) The indices_shape[-1] should have a value between 1 (inclusive) and rank r-b (inclusive)
14
+
8
- 3) All values in indices are expected to be within bounds [-s, s-1] along axis of size s (i.e.) -data_shape[i] <= indices[...,i] <= data_shape[i] - 1.
15
+ 5) All values in indices are expected to be within bounds [-s, s-1] along axis of size s (i.e.) -data_shape[i] <= indices[...,i] <= data_shape[i] - 1.
9
16
  It is an error if any of the index values are out of bounds.
10
17
  The output is computed as follows:
11
18
  The output tensor is obtained by mapping each index-tuple in the indices tensor to the corresponding slice of the input data.
12
- 1) If indices_shape[-1] > r => error condition
19
+ 1) If indices_shape[-1] > r-b => error condition
13
- 2) If indices_shape[-1] == r, since the rank of indices is q, indices can be thought of as a (q-1)-dimensional tensor
20
+ 2) If indices_shape[-1] == r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensors
14
- containing 1-D tensors of dimension r. Let us think of each such r ranked tensor as indices_slice.
21
+ containing 1-D tensors of dimension r-b, where N is an integer equals to the product of 1 and all the elements in the batch dimensions
15
- Each *scalar value* corresponding to data[indices_slice] is filled into the corresponding location of the (q-1)-dimensional tensor
22
+ of the indices_shape. Let us think of each such r-b ranked tensor as indices_slice. Each *scalar value* corresponding to data[0:b-1,indices_slice]
16
- to form the output tensor (Example 1 below)
23
+ is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Example 1 below)
17
- 3) If indices_shape[-1] < r, since the rank of indices is q, indices can be thought of as a (q-1)-dimensional tensor
24
+ 3) If indices_shape[-1] < r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensor
18
- containing 1-D tensors of dimension < r. Let us think of each such tensors as indices_slice.
25
+ containing 1-D tensors of dimension < r-b. Let us think of each such tensors as indices_slice. Each *tensor slice* corresponding
19
- Each *tensor slice* corresponding to data[indices_slice , :] is filled into the corresponding location of the (q-1)-dimensional tensor
26
+ to data[0:b-1, indices_slice , :] is filled into the corresponding location of the (q-b-1)-dimensional tensor
20
- to form the output tensor (Examples 2, 3, and 4 below)
27
+ to form the output tensor (Examples 2, 3, 4 and 5 below)
21
28
  This operator is the inverse of ScatterND.
22
29
  Example 1
30
+
31
+ batch_dims = 0
23
32
  data = [[0,1],[2,3]] # data_shape = [2, 2]
24
33
  indices = [[0,0],[1,1]] # indices_shape = [2, 2]
25
34
  output = [0,3] # output_shape = [2]
26
35
  Example 2
36
+ batch_dims = 0
37
+
27
38
  data = [[0,1],[2,3]] # data_shape = [2, 2]
28
39
  indices = [[1],[0]] # indices_shape = [2, 1]
29
40
  output = [[2,3],[0,1]] # output_shape = [2, 2]
30
41
  Example 3
42
+
43
+ batch_dims = 0
31
44
  data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
32
45
  indices = [[0,1],[1,0]] # indices_shape = [2, 2]
33
46
  output = [[2,3],[4,5]] # output_shape = [2, 2]
34
47
  Example 4
48
+ batch_dims = 0
49
+
35
50
  data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
36
51
  indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2]
37
52
  output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2]
53
+
54
+ Example 5
55
+
56
+ batch_dims = 1
57
+
58
+ data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
59
+
60
+ indices = [[1],[0]] # indices_shape = [2, 1]
61
+
62
+ output = [[2,3],[4,5]] # output_shape = [2, 2]
63
+
64
+ **Attributes**
65
+
66
+ * **batch_dims**:
67
+ The number of batch dimensions. The gather of indexing starts from
68
+ dimension of data[batch_dims:]
38
69
  **Inputs**
39
70
  * **data** (heterogeneous) - **T**:
40
71
  Tensor of rank r >= 1.
41
72
  * **indices** (heterogeneous) - **tensor(int64)**:
42
73
  Tensor of rank q >= 1. All index values are expected to be within
43
74
  bounds [-s, s-1] along axis of size s. It is an error if any of the
44
75
  index values are out of bounds.
45
76
  **Outputs**
46
77
  * **output** (heterogeneous) - **T**:
47
78
  Tensor of rank q + r - indices_shape[-1] - 1.
48
79
  **Type Constraints**
49
80
  * **T** in (
50
81
  tensor(bool),
51
82
  tensor(complex128),
52
83
  tensor(complex64),
53
84
  tensor(double),
54
85
  tensor(float),
55
86
  tensor(float16),
56
87
  tensor(int16),
57
88
  tensor(int32),
58
89
  tensor(int64),
59
90
  tensor(int8),
60
91
  tensor(string),
61
92
  tensor(uint16),
62
93
  tensor(uint32),
63
94
  tensor(uint64),
64
95
  tensor(uint8)
65
96
  ):
66
97
  Constrain input and output types to any tensor type.