ftorch_optim Module

Optimisers module for FTorch.

License FTorch is released under an MIT license. See the LICENSE file for details.

Uses

Help

Graph Key

Nodes of different colours represent the following:

Solid arrows point from a submodule to the (sub)module which it is descended from. Dashed arrows point from a module or program unit to modules which it uses.

Used by

Help

Graph Key

Nodes of different colours represent the following:

Solid arrows point from a submodule to the (sub)module which it is descended from. Dashed arrows point from a module or program unit to modules which it uses.

Derived Types

type, public :: torch_optim

Type for holding a torch optimizer.

Components

Type	Visibility	Attributes		Name		Initial
type(c_ptr),	public		::	p	=	c_null_ptr	pointer to the optimizer in memory

Finalizations Procedures

final :: torch_optim_delete

Type-Bound Procedures

procedure, public :: step => torch_optim_step
procedure, public :: zero_grad => torch_optim_zero_grad

Subroutines

public subroutine torch_optim_Adam(optim, parameters, learning_rate, beta_1, beta_2, eps, weight_decay, amsgrad)

Create an Adam optimizer

Arguments

Type	Intent	Optional	Attributes		Name
type(torch_optim),	intent(out)			::	optim	Optimizer we are creating
type(torch_tensor),	intent(in),		dimension(:)	::	parameters	Array of parameter tensors
real(kind=real64),	intent(in),	optional		::	learning_rate	learning rate for the optimization algorithm (default: 0.001)
real(kind=real64),	intent(in),	optional		::	beta_1	beta 1 for the optimization algorithm (default: 0.9)
real(kind=real64),	intent(in),	optional		::	beta_2	beta 2 for the optimization algorithm (default: 0.999)
real(kind=real64),	intent(in),	optional		::	eps	eps for the optimization algorithm (default: 1.0e-8)
real(kind=real64),	intent(in),	optional		::	weight_decay	weight_decay for the optimization algorithm (default: 0.0)
logical,	intent(in),	optional		::	amsgrad	enable AMSGrad variant (default: .false.)

public subroutine torch_optim_AdamW(optim, parameters, learning_rate, beta_1, beta_2, eps, weight_decay, amsgrad)

Create an AdamW optimizer

Arguments

Type	Intent	Optional	Attributes		Name
type(torch_optim),	intent(out)			::	optim	Optimizer we are creating
type(torch_tensor),	intent(in),		dimension(:)	::	parameters	Array of parameter tensors
real(kind=real64),	intent(in),	optional		::	learning_rate	learning rate for the optimization algorithm (default: 0.001)
real(kind=real64),	intent(in),	optional		::	beta_1	beta 1 for the optimization algorithm (default: 0.9)
real(kind=real64),	intent(in),	optional		::	beta_2	beta 2 for the optimization algorithm (default: 0.999)
real(kind=real64),	intent(in),	optional		::	eps	eps for the optimization algorithm (default: 1.0e-8)
real(kind=real64),	intent(in),	optional		::	weight_decay	weight_decay for the optimization algorithm (default: 0.01)
logical,	intent(in),	optional		::	amsgrad	enable AMSGrad variant (default: .false.)

public subroutine torch_optim_SGD(optim, parameters, learning_rate, momentum, weight_decay, dampening, nesterov)

Create an SGD optimizer

Arguments

Type	Intent	Optional	Attributes		Name
type(torch_optim),	intent(out)			::	optim	Optimizer we are creating
type(torch_tensor),	intent(in),		dimension(:)	::	parameters	Array of parameter tensors
real(kind=real64),	intent(in),	optional		::	learning_rate	learning rate for the optimization algorithm (default: 0.001)
real(kind=real64),	intent(in),	optional		::	momentum	momentum for the optimization algorithm (default: 0.0)
real(kind=real64),	intent(in),	optional		::	weight_decay	weight_decay for the optimization algorithm (default: 0.0)
real(kind=real64),	intent(in),	optional		::	dampening	dampening for the optimization algorithm (default: 0.0)
logical,	intent(in),	optional		::	nesterov	enable Nesterov momentum. Only applicable when momentum is non-zero. (default: .false.)

public subroutine torch_optim_delete(optim)

Deallocate a Torch optimizer

Arguments

Type	Intent	Optional	Attributes		Name
type(torch_optim),	intent(inout)			::	optim	Optimizer to deallocate

public subroutine torch_optim_step(optim)

Step a Torch optimizer

Arguments

Type	Intent	Optional	Attributes		Name
class(torch_optim),	intent(in)			::	optim	Optimizer to step

public subroutine torch_optim_zero_grad(optim)

Zero Gradients on tensors associated with a Torch optimizer

Arguments

Type	Intent	Optional	Attributes		Name
class(torch_optim),	intent(in)			::	optim	Optimizer to zero gradients for