fdapace (version 0.6.0)

WFDA: Time-Warping in Functional Data Analysis: Pairwise curve synchronization for functional data

Description

Time-Warping in Functional Data Analysis: Pairwise curve synchronization for functional data

Usage

WFDA(Ly, Lt, optns = list())

Value

A list containing the following fields:

optns

Control options used.

lambda

Penalty parameter used.

aligned

Aligned curves evaluated at time 't'

h

Warping functions for 't'

hInv

Inverse warping functions evaluated at 't'

costs

The mean cost associated with each curve

timing

The time required by time-warping.

Arguments

Ly

A list of n vectors containing the observed values for each individual.

Lt

A list of n vectors containing the observation time points for each individual corresponding to y. Each vector should be sorted in ascending order.

optns

A list of options control parameters specified by list(name=value). See 'Details'.

Details

WFDA uses a pairwise warping method to obtain the desired alignment (registration) of the random trajectories. The data has to be regular. The routine returns the aligned curves and the associated warping function.

Available control options are

choice

Choice of estimating the warping functions ('weighted' or 'truncated'). If 'weighted' then weighted averages of pairwise warping functions are computed; the weighting is based on the inverse pairwise distances. If 'truncated' the pairs with the top 10% largest distances are truncated and the simple average of the remaining pairwise distances are used - default: 'truncated'

subsetProp

Pairwise warping functions are determined by using a subset of the whole sample; numeric (0,1] - default: 0.50

lambda

Penalty parameter used for estimating pairwise warping functions; numeric - default : V*10^-4, where V is the average L2 norm of y-mean(y).

nknots

Number of knots used for estimating the piece-wise linear pairwise warping functions; numeric - default: 2

isPWL

Indicator if the resulting warping functions should piece-wise linear, if FALSE 'nknots' is ignored and the resulting warping functions are simply monotonic; logical - default: TRUE (significantly larger computation time.)

seed

Random seed for the selection of the subset of warping functions; numeric - default: 666

verbose

Indicator if the progress of the pairwise warping procedure should be displayed; logical - default: FALSE

References

Tang, R. and Müller, H.G. (2008). "Pairwise curve synchronization for functional data." Biometrika 95, 875-889

Tang, R. and Müller, H.G. (2009) "Time-synchronized clustering of gene expression trajectories." Biostatistics 10, 32-45

Examples

N = 44;
eps = 0.123;
M = 41;
set.seed(123) 
Tfinal = 3
me <- function(t) exp(-Tfinal*(((t/Tfinal^2)-0.5))^2);
T = seq(0,Tfinal,length.out = M) 
recondingTimesMat = matrix(nrow = N, ncol = M)
yMat = matrix(nrow = N, ncol = M)

for (i in 1:N){
  peak = runif(min = 0.2,max =  0.8,1) * Tfinal 
  recondingTimesMat[i,] = sort( unique(c( seq(0.0 , peak, length.out = round((M+1)/2)),
                            seq( peak, Tfinal, length.out = round((M+1)/2))))) * Tfinal
  yMat[i,] = me(recondingTimesMat[i,]) * rnorm(1, mean=4.0, sd=  eps)
                                       + rnorm(M, mean=0.0, sd=  eps) 
}

Y = as.list(as.data.frame(t(yMat)))
X = rep(list(T),N)
 
sss =  WFDA(Ly = Y, Lt = X, list( choice = 'weighted' ))
op <- par(mfrow=c(1,2))
matplot(x= T, t(yMat), t='l', main = 'Raw', ylab = 'Y'); grid()
matplot(x= T, t(sss$aligned), t='l', main = 'Aligned', ylab='Y'); grid() 
par(op)