create_dl_layer_gather🔗
Short description🔗
create_dl_layer_gather — Create a gather layer.
Signature🔗
create_dl_layer_gather( dl_layer DLLayerInput, number Indices, string LayerName, number Axis, attribute.name GenParamName, attribute.value GenParamValue, out dl_layer DLLayerGather )
Description🔗
The operator create_dl_layer_gather creates a gather layer that
extracts slices from the input tensor DLLayerInput along a
specified axis, according to the indices provided.
The handle of the gather layer is returned in DLLayerGather.
The parameter DLLayerInput determines the feeding input layer and
expects the layer handle as its value.
The parameter Indices defines the indices to be gathered from the
axis. The indices must be either of rank 0 or rank 1, i.e., a scalar or a
list. The indices input can be a layer handle (Indices) or a tuple
of integers.
If Indices is given as a tuple of integers, internally a constant
input layer is created.
The axis along which the gathering is performed is specified by the parameter
Axis. Axis can either be specified by a string
('width', 'height', 'depth' or 'batch') or an
integer, where 0 corresponds to the batch size, 1 to depth, 2 to height and 3
to width. Negative values for Axis indicate counting dimensions from
the back. The accepted range is [-4,3].
The parameter LayerName sets an individual layer name.
Note that if creating a model using create_dl_model, each layer
of the created network must have a unique name.
Gather semantics
Given a 4-dimensional input tensor DLLayerInput of shape
\((N, C, H, W)\) and an indices tensor or tuple,
the gather layer selects elements from the Axis dimension indexed
by elements of Indices.
If the indices are given as a tuple
\((i_{0}, \dots, i_{k})\)
and
Axis \(=a\), the output tensor is constructed by replacing dimension
\(a\) with the indices dimensions, resulting in an output of shape
\((d_{0}, \dots, d_{a-1}, k+1, d_{a+1}, \dots, d_{r-1})\)
All index values in Indices must be within the bounds
\([-s, s-1]\), where \(s\) is the size of the Axis
dimension. Negative indices are interpreted by adding \(s\), i.e.,
'index'\(=\)'index'\(+ s\) if 'index'\(< 0\).
Index Processing
-
Adjusting Axis Negative values for
Axisare made non-negative by adding the number of dimensions \(r\) of the input tensor:\[\begin{eqnarray*} \textrm{axis} \gets \textrm{axis} + r \qquad \textrm{if } \textrm{axis} < 0 \end{eqnarray*}\] -
Clamping Indices Negative values in
Indicesare adjusted by adding the size \(s\) of the corresponding axis:\[\begin{eqnarray*} \textrm{indices}[j] \gets \textrm{indices}[j] + s \qquad \textrm{if } \textrm{indices}[j] < 0 \end{eqnarray*}\]All values in
Indicesmust satisfy\[\begin{eqnarray*} 0 \leq \textrm{indices}[j] < s \end{eqnarray*}\]Otherwise, an error will be raised.
The following generic parameters GenParamName and the corresponding
values GenParamValue are supported:
-
'is_inference_output': Determines whether
apply_dl_modelwill include the output of this layer in the dictionaryDLResultBatcheven without specifying this layer inOutputs('true') or not ('false').List of values: 'true', 'false'
Default: 'false'
Certain parameters of layers created using this operator
create_dl_layer_gather can be set and retrieved using further operators.
The following tables give an overview, which parameters can be set
using set_dl_model_layer_param and which ones can be retrieved using
get_dl_model_layer_param or get_dl_layer_param.
Note, the operators set_dl_model_layer_param and
get_dl_model_layer_param require a model created by create_dl_model.
| Layer Parameters | set |
get |
|---|---|---|
'input_layer' (DLLayerInput,Indices) |
x |
|
'name' (LayerName) |
x |
x |
'axis' (Axis) |
x |
|
'output_layer' (DLLayerGather) |
x |
|
| 'shape' | x |
|
| 'type' | x |
| Generic Layer Parameters | set |
get |
|---|---|---|
| 'is_inference_output' | x |
x |
| 'num_trainable_params' | x |
Execution information🔗
Execution information
-
Multithreading type: reentrant (runs in parallel with non-exclusive operators).
-
Multithreading scope: global (may be called from any thread).
-
Processed without parallelization.
Parameters🔗
DLLayerInput (input_control) dl_layer → (handle)
Feeding data input layer.
Default: 'DLLayerInput'
Indices (input_control) number(-array) → (integer / handle)
Indices for the gather operation. This input can be a layer or tuple of integers.
LayerName (input_control) string → (string)
Name of the output layer.
Axis (input_control) number → (integer / string)
Axis along which to gather.
Default: 0
Suggested values: 0, 'batch', 1, 'depth', 2, 'height', 3, 'width'
GenParamName (input_control) attribute.name(-array) → (string)
Generic input parameter names.
Default: []
List of values: 'is_inference_output'
GenParamValue (input_control) attribute.value(-array) → (string / integer / real)
Generic input parameter values.
Default: []
Suggested values: 'true', 'false'
DLLayerGather (output_control) dl_layer → (handle)
Output layer.
Module🔗
Deep Learning Professional