UGParameterEstimator.linesearches package

Submodules

UGParameterEstimator.linesearches.lineSearches module

class UGParameterEstimator.linesearches.lineSearches.BacktrackingLineSearch(evaluator)

Bases: LineSearch

An serial (as in not parallelized) version of a backtracking line search, halving the step size if no suitable step size according to the first wolfe conditions is found.

c = 0.001
doLineSearch(stepdirection, guess, target, J, r, result)

Executes the line search along a given direction

Parameters

  • stepdirection (numpy array in parameter space) – the direction to search

  • guess (numpy array in parameter space) – the initial guess to do the step from

  • target (Evaluation) – the target of the calibration process

  • J (numpy array PxP) – the jacobian of the cost function in relation to the parameters

  • r (numpy array in measurement space) – the last residual vector

  • result – a tuple of final best guess, and overall

lowest residual value. Or (None, None), if an error occurred. :type result: tuple (numpy array, scalar)

max_iterations = 15
rho = 0.5
class UGParameterEstimator.linesearches.lineSearches.LineSearch(evaluator)

Bases: ABC

Base class for all line searches describing the interface and providing a helper function

abstract doLineSearch(stepdirection, guess, target, J, r, result)

Executes the line search along a given direction

Parameters

  • stepdirection (numpy array in parameter space) – the direction to search

  • guess (numpy array in parameter space) – the initial guess to do the step from

  • target (Evaluation) – the target of the calibration process

  • J (numpy array PxP) – the jacobian of the cost function in relation to the parameters

  • r (numpy array in measurement space) – the last residual vector

  • result – a tuple of final best guess, and overall

lowest residual value. Or (None, None), if an error occurred. :type result: tuple (numpy array, scalar)

measurementToNumpyArrayConverter(evaluations, target)

Helper function to convert an array of Evaluation. Each evaluation will be converted and interpolated using it’s getNumpyArrayLike method. None-values or Errors will be converted to None.

Parameters

  • evaluations (list of Evaluations) – the evaluations to convert

  • target – Evaluation describing the format/time steps

each evaluation should be converted/interpolated to :type target: Evaluation :return: the results of the covnertions :rtype: list of numpy arrays

class UGParameterEstimator.linesearches.lineSearches.LinearParallelLineSearch(evaluator, max_iterations=3, parallel_evaluations=10)

Bases: LineSearch

A line search executing multiple, uniformly distributed line search steps in parallel.

Multiple iterations, controlled by max_iterations, are performed sequentially, each with a smaller search window around the smallest window found in the last iteration.

If a suitable step size was found, as defined by the first wolfe condition with c = 1e-3, the search is stopped IF this value was not at the bounds of the search window. If it is, the search is continued in hope of a better step size.

c = 0.001
doLineSearch(stepdirection, guess, target, J, r, result)

Executes the line search along a given direction

Parameters

  • stepdirection (numpy array in parameter space) – the direction to search

  • guess (numpy array in parameter space) – the initial guess to do the step from

  • target (Evaluation) – the target of the calibration process

  • J (numpy array PxP) – the jacobian of the cost function in relation to the parameters

  • r (numpy array in measurement space) – the last residual vector

  • result – a tuple of final best guess, and overall

lowest residual value. Or (None, None), if an error occurred. :type result: tuple (numpy array, scalar)

max_iterations = 1
parallel_evaluations = 10
class UGParameterEstimator.linesearches.lineSearches.LogarithmicParallelLineSearch(evaluator, max_iterations=2, size=5, parallel_evaluations=10)

Bases: LineSearch

A parallel implementation of a line search, with the search values of the form 1/2^{-i} for multiple values of i in parallel.

If no suitable value (according to the first wolfe condition with c = 1e-3) is found, the next iteration is executed, with higher values for i.

c = 0.001
doLineSearch(stepdirection, guess, target, J, r, result)

Executes the line search along a given direction

Parameters

  • stepdirection (numpy array in parameter space) – the direction to search

  • guess (numpy array in parameter space) – the initial guess to do the step from

  • target (Evaluation) – the target of the calibration process

  • J (numpy array PxP) – the jacobian of the cost function in relation to the parameters

  • r (numpy array in measurement space) – the last residual vector

  • result – a tuple of final best guess, and overall

lowest residual value. Or (None, None), if an error occurred. :type result: tuple (numpy array, scalar)

highest_power = 0
max_iterations = 2
parallel_evaluations = 10
size = 5

Module contents