registerIterated: Registers a sample of curves to its weighted average, in an...

In snandi/Registration: Contains functions for registration algorithms, simulation and testing

Description

Uses the fda::register.fd function to register a sample of curves, after smoothing them, iteratively. Checks for outliers before estimating the weighted average. The weights are estimated based on distance from a L1-median robustX::L1median

Usage

registerIterated(
  dataToRegister, 
  Lambdas_ConstrainedWarping = c(0.001, 0.0001, 0.00005), 
  LambdaDefault = 0.00001,
  abscissaFrom, 
  abscissaTo, 
  abscissaIncrement, 
  basisOrder = 5, 
  basisBreakFreq = 3, 
  Lambdas_Roughness = exp(-5:5), 
  outlierTrimPct = 0.2, 
  N_Bootstrap_for_Outlier = 500,
  RE_REGISTER = FALSE,
  Convergence_Threshold = 1e-05,
  SimMeanDiff_Threshold = 0.0001, 
  MinSimilarityThreshold = 0.15, 
  MAX_ITERATION = 6
)

Arguments

dataToRegister	A matrix of curves to register. Each column is a new curve.
Lambdas_ConstrainedWarping	The roughness penalty for estimating the warping function. Smaller values will allow more undulations in the warping function.
LambdaDefault	The default roughness penalty, set to 0.00001. The λ is set to this value if the function iterates longer than the list of λ provided in Lambdas_ConstrainedWarping
abscissaFrom	Mininum value of abscissa
abscissaTo	Maximum value of abscissa
abscissaIncrement	Increment of abscissa
basisOrder	Order of B-spline basis functions
basisBreakFreq	Frequency of knots of basis functions
Lambdas_Roughness	Roughness penalty for smoothing using B-splines
outlierTrimPct	Percentage to trim when detecting outliers
N_Bootstrap_for_Outlier	Number of bootstrap samples to draw, to estimate outlier curves
RE_REGISTER	Re-register the curves from previous iterations, or register the original noisy curves, at every iteration, to the updated consensus/template
Convergence_Threshold	The value for argument conv in function register.fd
SimMeanDiff_Threshold	Defaults to 0.001, a criterion to stop the iterations
MinSimilarityThreshold	Minimum similarity between registered curve and template to estimate the final consensus
MAX_ITERATION	Maximum number of iterations

Value

dataToRegister	Original dataset to register
Regfd_Final	Final registered data object. Output of register.fd
registeredCurves	A matrix of registered curves extracted from the registered object
registeredCurves.D1	A matrix of first derivatives of registered curves extracted from the registered object
FinalConsensus	Final consensus of registered curves
curvesForConsensus	Names of curves used to estimate the consensus. This excludes the detected outliers
registeredCurvesAll	A matrix of all registered curves, including outliers
registeredCurvesAll.D1	A matrix of first derivatives of all registered curves, including outliers
Sim_toTemplate	A matrix of similarities of curves to template

Author(s)

Subhrangshu Nandi; [e-mail] or [e-mail]

See Also

register.fd, registerSingleIter

Examples

data( growth, package = 'fda' )
Mat1 <- growth[['hgtm']]
Arguments <- growth[['age']]

#Mat1Registered <- registerIterated(
#  dataToRegister = Mat1, 
#  Lambdas_ConstrainedWarping = c(0.001, 0.0001, 0.00005), 
#  LambdaDefault = 0.00001,
#  abscissaFrom = 1, 
#  abscissaTo = 16, 
#  abscissaIncrement = 0.5, 
#  basisOrder = 5, 
#  basisBreakFreq = 3, 
#  Lambdas_Roughness = exp(-5:5), 
#  outlierTrimPct = 0.05, 
#  N_Bootstrap_for_Outlier = 500,
#  RE_REGISTER = FALSE, 
#  Convergence_Threshold = 1e-05,
#  SimMeanDiff_Threshold = 0.001, 
#  MinSimilarityThreshold = 0.25, 
#  MAX_ITERATION = 4
#)

snandi/Registration documentation built on May 30, 2019, 5:04 a.m.