Package linear_operators :: Package iterative :: Module linesearch'

Module linesearch'


Line searches: find minimum of a multivariate function. 

Optionnaly depends on scipy.optimize for some line searches.

Available:

- optimal step (exact minimum if Criterion is quadratic (only Norm2
  norms))

- Backtracking : starts with optimal steps and reduces step until
  criterion decreases.

if scipy.optimize is in PYTHONPATH:

- LineSearch, LineSearchArmijo, LineSearchWolfe1; LineSearchWolfe2

Classes

[hide private]

Backtracking

LineSearch
Wraps scipy.optimize.linesearch.line_search

LineSearchArmijo
Wraps scipy.optimize.linesearch.line_search_armijo.

LineSearchWolfe1
Wraps scipy.optimize.linesearch.line_search_wolfe1

LineSearchWolfe2
Wraps scipy.optimize.linesearch.line_search_wolfe2

Functions

[hide private]

optimal_step(algo)
Finds quadratic optimal step of a criterion.

source code

Variables

[hide private]

__package__ = 'linear_operators.iterative'

default_backtracking = <linear_operators.iterative.linesearch....

Function Details

[hide private]

optimal_step(algo)

source code


Finds quadratic optimal step of a criterion.

Arguments
----------

algo: Algoritm instance with the following attributes:
  current_descent, current_gradient, criterion. The criterion
  attribute should be a Criterion instance with the following
  attributes: model, priors, hypers, norms.

Returns
-------
a: float
  The optimal step.

Variables Details

[hide private]

default_backtracking

Value:

<linear_operators.iterative.linesearch.Backtracking object at 0x28b269
0>