fnn_activation_linear Module Reference

Module dedicated to the class linearactivation. More...

Data Types
type	linearactivation
	Base class for all activation functions. Implements a linear activation function. More...

Functions/Subroutines
type(linearactivation) function, public	construct_linear_activation (self_size, batch_size)
	Constructor for class linearactivation. More...
subroutine	linear_tofile (self, unit_num)
	Implements linearactivation::tofile. More...
subroutine	linear_apply_forward (self, member, z, y)
	Implements linearactivation::apply_forward. More...
subroutine	linear_apply_tangent_linear (self, member, dz, dy)
	Implements linearactivation::apply_tangent_linear. More...
subroutine	linear_apply_adjoint (self, member, dy, dz)
	Implements linearactivation::apply_adjoint. More...

Detailed Description

Module dedicated to the class linearactivation.

Function/Subroutine Documentation

construct_linear_activation()

type(linearactivation) function, public fnn_activation_linear::construct_linear_activation	(	integer(ik), intent(in)	self_size,
		integer(ik), intent(in)	batch_size
	)

Constructor for class linearactivation.

Parameters

[in]	self_size	The value for linearactivation::self_size.
[in]	batch_size	The value for linearactivation::batch_size.

Returns

The constructed activation function.

linear_apply_adjoint()

subroutine fnn_activation_linear::linear_apply_adjoint ( class(linearactivation), intent(in)  self, integer(ik), intent(in)  member, real(rk), dimension(:), intent(in)  dy, real(rk), dimension(:), intent(out)  dz  )

private

Implements linearactivation::apply_adjoint.

Applies the adjoint of the activation function.

The adjoint operator reads

Note

In principle, this method should only be called after linearactivation::apply_forward, where the linearisation is computed.

This method supports inplace operations: it can be called with dz = dy.

Parameters

[in]	self	The activation function.
[in]	member	The index inside the batch.
[in]	dy	The input of the adjoint operator.
[out]	dz	The output of the adjoint operator.

linear_apply_forward()

subroutine fnn_activation_linear::linear_apply_forward ( class(linearactivation), intent(inout)  self, integer(ik), intent(in)  member, real(rk), dimension(:), intent(in)  z, real(rk), dimension(:), intent(out)  y  )

private

Implements linearactivation::apply_forward.

Applies and linearises the activation function.

The activation function reads

and the associated linearisation reads

which is trivial and does not require any operation.

Note

Input parameter member should be less than linearactivation::batch_size.

This method supports inplace operations: it can be called with y = z.

The intent for self is declared inout instead of in because, for certain subclasses (e.g. fnn_activation_tanh::tanhactivation) the linearisation is stored inside the activation function.

Parameters

[in,out]	self	The activation function.
[in]	member	The index inside the batch.
[in]	z	The input of the activation function.
[out]	y	The output of the activation function.

linear_apply_tangent_linear()

subroutine fnn_activation_linear::linear_apply_tangent_linear ( class(linearactivation), intent(in)  self, integer(ik), intent(in)  member, real(rk), dimension(:), intent(in)  dz, real(rk), dimension(:), intent(out)  dy  )

private

Implements linearactivation::apply_tangent_linear.

Applies the TL of the activation function.

The TL operator reads

Note

In principle, this method should only be called after linearactivation::apply_forward, where the linearisation is computed.

This method supports inplace operations: it can be called with dy = dz.

Parameters

[in]	self	The activation function.
[in]	member	The index inside the batch.
[in]	dz	The input of the TL operator.
[out]	dy	The output of the TL operator.

linear_tofile()

subroutine fnn_activation_linear::linear_tofile ( class(linearactivation), intent(in)  self, integer(ik), intent(in)  unit_num  )

private

Implements linearactivation::tofile.

Saves the activation function.

Parameters

[in]	self	The activation function to save.
[in]	unit_num	The unit number for the write statement.