Compute Library
 21.02
FullyConnectedLayerInfo Struct Reference

Fully connected layer info. More...

#include <Types.h>

Collaboration diagram for FullyConnectedLayerInfo:
[legend]

Public Member Functions

FullyConnectedLayerInfoset_weights_trained_layout (DataLayout layout)
 Sets the weights trained data layout. More...
 
FullyConnectedLayerInfoset_transpose_weights (bool should_transpose_weights)
 Sets the transpose weights flag. More...
 

Data Fields

DataLayout weights_trained_layout { DataLayout::NCHW }
 Layout that the weights have been trained with. More...
 
bool transpose_weights { true }
 Transpose weights if true. More...
 
bool are_weights_reshaped { false }
 Reshape the weights tensor if false. More...
 
bool retain_internal_weights { false }
 Retain internal reshaped weights. More...
 
bool fp_mixed_precision { false }
 Use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy. More...
 
ActivationLayerInfo activation_info {}
 Fused activation to apply after the matrix multiplication. More...
 

Detailed Description

Fully connected layer info.

Definition at line 1613 of file Types.h.

Member Function Documentation

◆ set_transpose_weights()

FullyConnectedLayerInfo& set_transpose_weights ( bool  should_transpose_weights)
inline

Sets the transpose weights flag.

Parameters
[in]should_transpose_weightsBoolean flag indicating if weights should be transposed
Returns
Updated object

Definition at line 1639 of file Types.h.

1640  {
1641  transpose_weights = should_transpose_weights;
1642  return *this;
1643  }
bool transpose_weights
Transpose weights if true.
Definition: Types.h:1616

◆ set_weights_trained_layout()

FullyConnectedLayerInfo& set_weights_trained_layout ( DataLayout  layout)
inline

Sets the weights trained data layout.

Parameters
[in]layoutData layout that the weights were trained with
Returns
Updated object

Definition at line 1628 of file Types.h.

1629  {
1630  weights_trained_layout = layout;
1631  return *this;
1632  }
DataLayout weights_trained_layout
Layout that the weights have been trained with.
Definition: Types.h:1615

Field Documentation

◆ activation_info

ActivationLayerInfo activation_info {}

Fused activation to apply after the matrix multiplication.

Definition at line 1620 of file Types.h.

Referenced by NEFullyConnectedLayer::configure(), FullyConnectedLayerNode::set_fused_activation(), NEFullyConnectedLayer::validate(), and CLFullyConnectedLayer::validate().

◆ are_weights_reshaped

◆ fp_mixed_precision

bool fp_mixed_precision { false }

Use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy.

Definition at line 1619 of file Types.h.

◆ retain_internal_weights

bool retain_internal_weights { false }

◆ transpose_weights

◆ weights_trained_layout

DataLayout weights_trained_layout { DataLayout::NCHW }

Layout that the weights have been trained with.

Definition at line 1615 of file Types.h.

Referenced by NEFullyConnectedLayer::configure(), CLFullyConnectedLayer::configure(), NEFullyConnectedLayer::validate(), and CLFullyConnectedLayer::validate().


The documentation for this struct was generated from the following file: