def defaultParameters():
"""Default parameters that can subsequently be modified and passed to the constructor.
:param ITER_MAX: maximum number of minimization iterations before failure
:param mu: Morozov discrepancy principle scaling paramter
:param cg_reset: number of iterations before a conjugate gradient reset (0=default)
:param thetaMin: smallest acceptable goal improvment for linearized step
:param thetaMax: maximum attempted goal improvment for linearized step
:param kappaTrust: factor to shrink :data:`theta` by when linearized step fails
:param rhoLow: linearized step fails when goal improves by less than :data:`rhoLow`*:data:`theta`
:param rhoHigh: linearized step is very successful when goal improves by more than :data:`rhoHigh`*:data:`theta`
:param verbose: True if extra logging output is desired
"""
params = Parameters('InvIGN', ITER_MAX=200, mu=1.1, cg_reset=0, thetaMin=2**(-20), thetaMax=0.5,
kappaTrust=0.5, rhoLow=0.1, rhoHigh=0.5, verbose=False)
lsparams = linesearchHZ.LinesearchHZ.defaultParameters()
lsparams.sigma=0.1
lsparams.delta=0.05
lsparams.rename('linesearch')
params.add(lsparams)
linearparams = KrylovSolver.defaultParameters()
linearparams.mu=1.
linearparams.rename('linearsolver')
params.add(linearparams)
return params
def defaultParameters():
"""Parameters:
* ITER_MAX: maximum iteration count
* mu: scaling parameter for Morozov discrepency principle.
* cg_reset: reset conjugate gradient method after this many iterations. Zero for a default behaviour.
* steepest_descent: Use steepest descent rather than full NLCG
* linesearch: parameters to be passed to the linesearch algorithm
* verbose: print out extra output
"""
params = Parameters('InvNLCG', ITER_MAX=200, mu=1.1, cg_reset=0, deriv_eps=1, steepest_descent=False, verbose=False)
lsparams = linesearchHZ.LinesearchHZ.defaultParameters()
lsparams.sigma=0.1
lsparams.delta=0.05
lsparams.rename('linesearch')
params.add(lsparams)
return params
def defaultParameters():
"""Default parameters that can subsequently be modified and passed to the constructor.
:param ITER_MAX: maximum number of minimization iterations before failure
:param mu: Morozov discrepancy principle scaling paramter
:param cg_reset: number of iterations before a conjugate gradient reset (0=default)
:param thetaMin: smallest acceptable goal improvment for linearized step
:param thetaMax: maximum attempted goal improvment for linearized step
:param kappaTrust: factor to shrink :data:`theta` by when linearized step fails
:param rhoLow: linearized step fails when goal improves by less than :data:`rhoLow`*:data:`theta`
:param rhoHigh: linearized step is very successful when goal improves by more than :data:`rhoHigh`*:data:`theta`
:param verbose: True if extra logging output is desired
"""
params = Parameters('InvIGN',
ITER_MAX=200,
mu=1.1,
cg_reset=0,
thetaMin=2**(-20),
thetaMax=0.5,
kappaTrust=0.5,
rhoLow=0.1,
rhoHigh=0.5,
verbose=False)
lsparams = linesearchHZ.LinesearchHZ.defaultParameters()
lsparams.sigma = 0.1
lsparams.delta = 0.05
lsparams.rename('linesearch')
params.add(lsparams)
linearparams = KrylovSolver.defaultParameters()
linearparams.mu = 1.
linearparams.rename('linearsolver')
params.add(linearparams)
return params
def defaultParameters():
"""Parameters:
* ITER_MAX: maximum iteration count
* mu: scaling parameter for Morozov discrepency principle.
* cg_reset: reset conjugate gradient method after this many iterations. Zero for a default behaviour.
* steepest_descent: Use steepest descent rather than full NLCG
* linesearch: parameters to be passed to the linesearch algorithm
* verbose: print out extra output
"""
params = Parameters('InvNLCG',
ITER_MAX=200,
mu=1.1,
cg_reset=0,
deriv_eps=1,
steepest_descent=False,
verbose=False)
lsparams = linesearchHZ.LinesearchHZ.defaultParameters()
lsparams.sigma = 0.1
lsparams.delta = 0.05
lsparams.rename('linesearch')
params.add(lsparams)
return params
def defaultParameters():
return Parameters('KrylovSolver',
ITER_MAX=200,
mu=1.1,
cg_reset=0,
steepest_descent=False)