loco::FeatureDecoder Struct Reference

Describe how to build a tensor from a (convolution) feature map. More...

#include <FeatureCodec.h>

Public Member Functions
virtual	~FeatureDecoder ()=default
virtual TensorShape	shape (const FeatureShape &) const =0
virtual FeatureIndex	value (const TensorIndex &) const =0
virtual std::unique_ptr< FeatureDecoder >	clone (void) const =0

Detailed Description

Describe how to build a tensor from a (convolution) feature map.

Let us assume that "dec" is a feature decoder.

Given a feature map "inp" and its shape "inp.shape", "dec" builds a tensor "out" as follows:

for each valid tensor index (referred to as tensor_index below) for dec.shape(inp.shape) out.at(tensor_index) = inp.at(dec.value(tensor_index))

NOTE "inp" is a feature value and "out" is a tensor value in this example.

Definition at line 67 of file FeatureCodec.h.

Constructor & Destructor Documentation

~FeatureDecoder()

virtual loco::FeatureDecoder::~FeatureDecoder ( )

virtualdefault

Member Function Documentation

clone()

virtual std::unique_ptr< FeatureDecoder > loco::FeatureDecoder::clone ( void  ) const

pure virtual

Implemented in loco::PermutingDecoder< Domain::Feature >.

shape()

virtual TensorShape loco::FeatureDecoder::shape ( const FeatureShape &  ) const

pure virtual

Implemented in loco::PermutingDecoder< Domain::Feature >.

Referenced by RandomDataGenerator.RandomDataGenerator::_gen_float32(), RandomDataGenerator.RandomDataGenerator::_gen_int16(), and RandomDataGenerator.RandomDataGenerator::_gen_uint8().

value()

virtual FeatureIndex loco::FeatureDecoder::value ( const TensorIndex &  ) const

pure virtual

Implemented in loco::PermutingDecoder< Domain::Feature >.

---

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

• compiler/loco/include/loco/IR/FeatureCodec.h