|
|
@@ -9817,7 +9817,7 @@
|
|
|
"module": "ai.onnx",
|
|
|
"version": 13,
|
|
|
"support_level": "common",
|
|
|
- "description": "Given `data` tensor of rank r >= 1, and `indices` tensor of rank q, gather\nentries of the axis dimension of `data` (by default outer-most one as axis=0) indexed by `indices`, and concatenates\nthem in an output tensor of rank q + (r - 1).\n\naxis = 0 :\n\nLet\nk = indices[i_{0}, ..., i_{q-1}]\nThen\noutput[i_{0}, ..., i_{q-1}, j_{0}, ..., j_{r-2}] = input[k , j_{0}, ..., j_{r-2}]\n\n```\n data = [\n [1.0, 1.2],\n [2.3, 3.4],\n [4.5, 5.7],\n ]\n indices = [\n [0, 1],\n [1, 2],\n ]\n output = [\n [\n [1.0, 1.2],\n [2.3, 3.4],\n ],\n [\n [2.3, 3.4],\n [4.5, 5.7],\n ],\n ]\n```\naxis = 1 :\n\nLet\nk = indices[i_{0}, ..., i_{q-1}]\nThen\noutput[i_{0}, ..., i_{q-1}, j_{0}, ..., j_{r-2}] = input[j_{0}, k, j_{1}, ..., j_{r-2}]\n\n```\n data = [\n [1.0, 1.2, 1.9],\n [2.3, 3.4, 3.9],\n [4.5, 5.7, 5.9],\n ]\n indices = [\n [0, 2],\n ]\n axis = 1,\n output = [\n [[1.0, 1.9]],\n [[2.3, 3.9]],\n [[4.5, 5.9]],\n ]\n```\n",
|
|
|
+ "description": "Given `data` tensor of rank r >= 1, and `indices` tensor of rank q, gather\nentries of the axis dimension of `data` (by default outer-most one as axis=0) indexed by `indices`, and concatenates\nthem in an output tensor of rank q + (r - 1).\n\naxis = 0 :\n\nLet\nk = indices[i_{0}, ..., i_{q-1}]\nThen\noutput[i_{0}, ..., i_{q-1}, j_{0}, ..., j_{r-2}] = input[k , j_{0}, ..., j_{r-2}]\n\n```\n data = [\n [1.0, 1.2],\n [2.3, 3.4],\n [4.5, 5.7],\n ]\n indices = [\n [0, 1],\n [1, 2],\n ]\n output = [\n [\n [1.0, 1.2],\n [2.3, 3.4],\n ],\n [\n [2.3, 3.4],\n [4.5, 5.7],\n ],\n ]\n```\naxis = 1 :\n\nLet\nk = indices[i_{0}, ..., i_{q-1}]\nThen\noutput[j_{0}, i_{0}, ..., i_{q-1}, j_{1}, ..., j_{r-2}] = input[j_{0}, k, j_{1}, ..., j_{r-2}]\n\n```\n data = [\n [1.0, 1.2, 1.9],\n [2.3, 3.4, 3.9],\n [4.5, 5.7, 5.9],\n ]\n indices = [\n [0, 2],\n ]\n axis = 1,\n output = [\n [[1.0, 1.9]],\n [[2.3, 3.9]],\n [[4.5, 5.9]],\n ]\n```\n",
|
|
|
"attributes": [
|
|
|
{
|
|
|
"name": "axis",
|
|
|
@@ -29082,7 +29082,7 @@
|
|
|
"module": "ai.onnx",
|
|
|
"version": 13,
|
|
|
"support_level": "common",
|
|
|
- "description": "Produces a slice of the input tensor along multiple axes. Similar to numpy:\nhttps://numpy.org/doc/stable/user/basics.indexing.html?highlight=slice#slicing-and-striding\n\nSlice uses the `starts`, `ends`, `axes` and `steps` inputs to select a sub-tensor\nof its input `data` tensor.\n\nAn effective `start[i]`, `end[i]`, and `step[i]` must be computed for each `i`\nin `[0, ... r-1]` where `r = rank(input)` as follows:\n\nIf `axes` are omitted, they are set to `[0, ..., r-1]`.\nIf `steps` are omitted, they are set to `[1, ..., 1]` of length `len(starts)`\n\nThe effective values are initialized as `start[i] = 0`, `end[i] = dims[i]` where\n`dims` are the dimensions of `input` and `step[i] = `1.\n\nAll negative elements of `axes` are made non-negatve by adding `r` to them, where\n`r =rank(input)`.\n\nAll negative values in `starts[i]` and `ends[i]` have `dims[axes[i]]` added to them,\nwhere `dims` are the dimensions of `input`. Then `start[axes[i]]` is the adjusted\n`starts[i]` clamped into range of valid indices, i.e. `[0, dims[axes[i]]-1]`.\n\nThe clamping for the adjusted `ends[i]` depends on the sign of `steps[i]` and must\naccommodate copying 0 through `dims[axes[i]]` elements, so for positive stepping\n`end[axes[i]]` is clamped to `[0, dims[axes[i]]]`, while for negative stepping it\nis clamped to `[-1, ends[i]-1]`.\n\nFinally, `step[axes[i]] = steps[i]`.\n\nFor slicing to the end of a dimension with unknown size, it is recommended to pass\nin `INT_MAX` when slicing forward and 'INT_MIN' when slicing backward.\n\nExample 1:\n data = [\n [1, 2, 3, 4],\n [5, 6, 7, 8],\n ]\n axes = [0, 1]\n starts = [1, 0]\n ends = [2, 3]\n steps = [1, 2]\n result = [\n [5, 7],\n ]\nExample 2:\n data = [\n [1, 2, 3, 4],\n [5, 6, 7, 8],\n ]\n starts = [0, 1]\n ends = [-1, 1000]\n result = [\n [2, 3, 4],\n ]\n",
|
|
|
+ "description": "Produces a slice of the input tensor along multiple axes. Similar to numpy:\nhttps://numpy.org/doc/stable/user/basics.indexing.html?highlight=slice#slicing-and-striding\n\nSlice uses the `starts`, `ends`, `axes` and `steps` inputs to select a sub-tensor\nof its input `data` tensor.\n\nAn effective `start[i]`, `end[i]`, and `step[i]` must be computed for each `i`\nin `[0, ... r-1]` where `r = rank(input)` as follows:\n\nIf `axes` are omitted, they are set to `[0, ..., r-1]`.\nIf `steps` are omitted, they are set to `[1, ..., 1]` of length `len(starts)`\n\nThe effective values are initialized as `start[i] = 0`, `end[i] = dims[i]` where\n`dims` are the dimensions of `input` and `step[i] = `1.\n\nAll negative elements of `axes` are made non-negatve by adding `r` to them, where\n`r =rank(input)`.\n\nAll negative values in `starts[i]` and `ends[i]` have `dims[axes[i]]` added to them,\nwhere `dims` are the dimensions of `input`. Then `start[axes[i]]` is the adjusted\n`starts[i]` is clamped into the range `[0, dims[axes[i]]]` for positive stepping\nand `[0, dims[axes[i]]-1]` for negative stepping.\n\nThe clamping for the adjusted `ends[i]` depends on the sign of `steps[i]` and must\naccommodate copying 0 through `dims[axes[i]]` elements, so for positive stepping\n`end[axes[i]]` is clamped to `[0, dims[axes[i]]]`, while for negative stepping it\nis clamped to `[-1, dims[axes[i]]-1]`.\n\nFinally, `step[axes[i]] = steps[i]`.\n\nFor slicing to the end of a dimension with unknown size, it is recommended to pass\nin `INT_MAX` when slicing forward and 'INT_MIN' when slicing backward.\n\nExample 1:\n data = [\n [1, 2, 3, 4],\n [5, 6, 7, 8],\n ]\n axes = [0, 1]\n starts = [1, 0]\n ends = [2, 3]\n steps = [1, 2]\n result = [\n [5, 7],\n ]\nExample 2:\n data = [\n [1, 2, 3, 4],\n [5, 6, 7, 8],\n ]\n starts = [0, 1]\n ends = [-1, 1000]\n result = [\n [2, 3, 4],\n ]\n",
|
|
|
"inputs": [
|
|
|
{
|
|
|
"name": "data",
|
Lutz Roeder