utils::Function< T > Class Template Reference

Function node in the computational graph. More...

#include <autograd.hpp>

Inheritance diagram for utils::Function< T >:

Collaboration diagram for utils::Function< T >:

Public Member Functions
virtual	~Function ()=default
virtual Matrix< T >	forward (const std::vector< Variable< T > > &inputs)=0
	Forward pass computation.
virtual std::vector< Matrix< T > >	backward (const Matrix< T > &grad_output)=0
	Backward pass computation.
virtual void	save_for_backward (const std::vector< Matrix< T > > &tensors)
	Set saved tensors for backward pass.

Protected Attributes
std::vector< Matrix< T > >	saved_tensors_

Detailed Description

template<typename T>
class utils::Function< T >

Function node in the computational graph.

Definition at line 25 of file autograd.hpp.

Constructor & Destructor Documentation

~Function()

template<typename T >

virtual utils::Function< T >::~Function ( )

virtualdefault

Member Function Documentation

backward()

template<typename T >

virtual std::vector< Matrix< T > > utils::Function< T >::backward ( const Matrix< T > &  grad_output )

pure virtual

Backward pass computation.

Parameters

grad_output

Gradient from the output

Returns

Gradients with respect to inputs

Implemented in utils::AddFunction< T >, utils::SubFunction< T >, utils::MulFunction< T >, utils::DotFunction< T >, utils::TransposeFunction< T >, utils::SigmoidFunction< T >, and utils::SumFunction< T >.

forward()

template<typename T >

virtual Matrix< T > utils::Function< T >::forward ( const std::vector< Variable< T > > &  inputs )

pure virtual

Forward pass computation.

Parameters

inputs

Input variables

Returns

Output matrix

Implemented in utils::AddFunction< T >, utils::SubFunction< T >, utils::MulFunction< T >, utils::DotFunction< T >, utils::TransposeFunction< T >, utils::SigmoidFunction< T >, and utils::SumFunction< T >.

save_for_backward()

template<typename T >

virtual void utils::Function< T >::save_for_backward ( const std::vector< Matrix< T > > &  tensors )

inlinevirtual

Set saved tensors for backward pass.

Definition at line 46 of file autograd.hpp.

Member Data Documentation

saved_tensors_

template<typename T >

std::vector<Matrix<T> > utils::Function< T >::saved_tensors_

protected

Definition at line 51 of file autograd.hpp.

---

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

• include/utils/autograd.hpp