module onnxrt.ops_cpu.op_scatternd#

Inheritance diagram of mlprodict.onnxrt.ops_cpu.op_scatternd

Short summary#

module mlprodict.onnxrt.ops_cpu.op_scatternd

Runtime operator.

source on GitHub

Classes#

class

truncated documentation

ScatterND

ScatterND ========= ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates

Functions#

function

truncated documentation

_scatter_nd_impl

Properties#

property

truncated documentation

args_default

Returns the list of arguments as well as the list of parameters with the default values (close to the signature). …

args_default_modified

Returns the list of modified parameters.

args_mandatory

Returns the list of optional arguments.

args_optional

Returns the list of optional arguments.

atts_value

Returns all parameters in a dictionary.

Methods#

method

truncated documentation

__init__

_infer_shapes

_run

Documentation#

Runtime operator.

source on GitHub

class mlprodict.onnxrt.ops_cpu.op_scatternd.ScatterND(onnx_node, desc=None, **options)#

Bases: OpRun

ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates tensor of rank q + r - indices.shape[-1] - 1. The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data. Note that indices should not have duplicate entries. That is, two or more updates for the same index-location is not supported.

indices is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of indices.

indices is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into data.

Hence, k can be a value at most the rank of data. When k equals rank(data), each update entry specifies an update to a single element of the tensor. When k is less than rank(data) each update entry specifies an update to a slice of the tensor.

updates is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. The remaining dimensions of updates correspond to the dimensions of the replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, corresponding to the trailing (r-k) dimensions of data. Thus, the shape of updates must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation of shapes.

The output is calculated via the following equation:

output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):

output[indices[idx]] = updates[idx]

The order of iteration in the above loop is not specified. In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order.

reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. When reduction is set to “add”, output is calculated as follows:

output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):

output[indices[idx]] += updates[idx]

When reduction is set to “mul”, output is calculated as follows:

output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):

output[indices[idx]] *= updates[idx]

This operator is the inverse of GatherND.

Example 1: ``

data = [1, 2, 3, 4, 5, 6, 7, 8] indices = [[4], [3], [1], [7]] updates = [9, 10, 11, 12] output = [1, 11, 3, 10, 9, 6, 7, 12]

``

Example 2: ``

data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]],

[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]

indices = [[0], [2]] updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]],

[[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]]

output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]],

[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]

``

Attributes

  • reduction: Type of reduction to apply: none (default), add, mul. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the multiplication operation. Default value is namereductionsnonetypeSTRING (STRING)

Inputs

  • data (heterogeneous)T: Tensor of rank r >= 1.

  • indices (heterogeneous)tensor(int64): Tensor of rank q >= 1.

  • updates (heterogeneous)T: Tensor of rank q + r - indices_shape[-1] - 1.

Outputs

  • output (heterogeneous)T: Tensor of rank r >= 1.

Type Constraints

  • T tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128): Constrain input and output types to any tensor type.

Version

Onnx name: ScatterND

This version of the operator has been available since version 16.

Runtime implementation: ScatterND

__init__(onnx_node, desc=None, **options)#
_infer_shapes(data, indices, updates)#

Should be overwritten.

source on GitHub

_run(data, indices, updates, attributes=None, verbose=0, fLOG=None)#

Should be overwritten.

source on GitHub

mlprodict.onnxrt.ops_cpu.op_scatternd._scatter_nd_impl(data, indices, updates, reduction=b'none')#