PyDSTool.Toolbox.optimizers.step

Package step

Module containing every step use to lower a cost function

Steps :

GradientStep

compute a step based on the gradient of the function

CWConjugateGradient

Crowder-Wolfe conjugate gradient

DYConjugateGradient

Dai-Yuan conjugate gradient

DConjugateGradient

Dixon conjugate gradient

FRConjugateGradient

Fletcher-Reeves conjugate gradient

PRPConjugateGradient

Polak-Ribiere-Polyak conjugate gradient

FRPRPConjugateGradient

Fletcher-Reeves modified Polak-Ribiere-Polyak conjugate gradient

NewtonStep

computes a step based on the hessian and the gradient of the function

MarquardtStep

computes a step based on the Marquardt modified hessian and the gradient of the function

GoldsteinPriceStep

computes a step based on the Goldstein-Price Newton modification

GoldfeldStep

computes a step based on the Goldfeld Newton modification

DFPNewtonStep

computes a step based on the Davidson-Fletcher-Powell approximation of the hessian

PartialStep

decorator for another step but uses only part of this step

RestartPeriodicallyConjugateGradientStep

decorator for a conjugate gradient step that restarts the conjugation each n iterations

RestartNotOrthogonalConjugateGradientStep

decorator for a conjugate gradient step that restarts the conjugation if the last gradients are not orthogonal enough

LocalBruteForce1Dstep

local brute force step in 1D (experimental)

Submodules

[hide private]

PyDSTool.Toolbox.optimizers.step.conjugate_gradient_step: Computes the conjugate gradient steps for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.goldfeld_step: Computes Goldfeld step for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.goldstein_price_step: Computes Goldstein-Price step for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.gradient_step: Computes a gradient step for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.local_brute_force_1dstep
PyDSTool.Toolbox.optimizers.step.marquardt_step: Computes a Marquardt step for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.newton_step: Computes a Newton step for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.partial_step: Computes a partial step for a specific function at a specific point, acting like a decorator for other steps
PyDSTool.Toolbox.optimizers.step.quasi_newton_step: Computes a quasi-Newton step for a specific function at a specific point
PyDSTool.Toolbox.optimizers.step.restart_conjugate_gradient: Restarts a conjugate gradient search by deleting the step key in the state dictionary

Classes

[hide private]

GradientStep
The simple gradient step

NewtonStep
The Newton step

MarquardtStep
The simple gradient step

GoldfeldStep
The Goldfeld step

GoldsteinPriceStep
The Goldstein-Price step

PartialStep
A partial step

RestartPeriodicallyConjugateGradientStep
A step decorator that periodically deletes the direction key in the dictionary so that the CG search can be restarted

RestartNotOrthogonalConjugateGradientStep
A step decorator that deletes the direction key in the dictionary if the last gradients are not orthogonal enough so that the CG search can be restarted

LocalBruteForce1DStep
Local brute force search for 1D parameter (sub-)space, making no use of gradient information.

Functions

[hide private]

DYConjugateGradientStep()
The Dai Yan conjugate gradient step Has good convergence capabilities (same as the FR-PRP gradient)

source code

FRPRPConjugateGradientStep()
The Fletcher-Reeves modified Polak-Ribiere-Polyak conjugate gradient step Can restart automatically and has the advantages of the PRP gradient and of the FR gradient

source code

PRPConjugateGradientStep()
The Polak-Ribiere-Polyak conjugate gradient step Can restart automatically, but needs an exact line search with a uniformely convex function to globally converge

source code

CWConjugateGradientStep()
The Crowder-Wolfe or Hestenes-Stiefel conjugate gradient step

source code

FRConjugateGradientStep()
The Fletcher Reeves conjugate gradient step Needs an exact line search for convergence or the strong Wolfe-Powell rules for an inexact line search

source code

DConjugateGradientStep()
The Dixon conjugate gradient step

source code

Variables

[hide private]

step__all__ = ['GradientStep', 'CWConjugateGradientStep', 'DYC...

Variables Details

[hide private]

stepall

Value:

['GradientStep',
 'CWConjugateGradientStep',
 'DYConjugateGradientStep',
 'DConjugateGradientStep',
 'FRConjugateGradientStep',
 'PRPConjugateGradientStep',
 'FRPRPConjugateGradientStep',
 'NewtonStep',
...