Performs composite time series outlier ensembling.

Performs composite time series outlier ensembling.

comp_tsout_ens(
  x,
  m1 = NULL,
  ncomp = 2,
  sds = 1,
  rept = 1,
  compr = 2,
  rat = 0.05,
  fast = TRUE
)

Arguments

x

A data frame or a matrix object containing a multivariate time series

m1

Variable indicating dimension reduction methods. Default is set to using all 4 methods: PCA, DOBIN, ICS and ICA.

ncomp

The number of components for each dimension reduction method. Default is set to 2.

sds

The random seed for generating a no-outlier time series.

rept

The number of repetitions for generating a no-outlier time series.

compr

To adjust for multiple testing, the results of the ensemble are compared with the results of a time series without outliers. If compr =1, a time series is simulated as in simulate_comp_ts without outliers. If compr = 2, the top outliers are removed from the outlier series and interpolated values are used for those time points. If compr = 3 both methods of simulation are used for comparison.

rat

A comparison is done with the outliers removed time series. The variable rat denotes the ratio of outliers to be removed as a proportion of the whole dataset for this comparison.

fast

For faster computation skip ICS decomposition method.

Value

A list with the following components:

outliers

The outliers detected from the multivariate ensemble after comparing with the comparison time series without outliers.

all

All the outliers detected from the multivariate ensemble.

outmat

A matrix with outlier scores organised by outlier method.

wts

The weights of the outlier detection methods.

pca_loadings

The basis vectors from PCA.

dobin_loadings

The basis vectors from DOBIN. See R package dobin for more details.

ics_loadings

The basis vectors from ICS. See R package ICS for more details.

ica_loadings

The basis vectors from Independent Component Analysis.

decomp_wts

Each decomposition method has several components. For example if ncomp=2, then there are 2 PC components, 2 DOBIn components, etc ... The weight of each component is given different and depends on the decomposition method. These weights are given in decomp_wts.

outmat4D

A 4D array with outlier scores organised by outlier method, decomposition method, components for each decomposition method and time.

comp_loadings

The unconstrained basis vectors on the simplex.

comp_coords

The unconstrained coordinates of the composite time series data.

Examples

if (FALSE) {
set.seed(100)
n <- 600
x <- sample(1:100, n, replace=TRUE)
x[25] <- 200
x[320] <- 300
x2 <- sample(1:100, n, replace=TRUE)
x3 <- sample(1:100, n, replace=TRUE)
x4 <- sample(1:100, n, replace=TRUE)
X <- cbind.data.frame(x, x2, x3, x4)
X <- X/rowSums(X)
out <- comp_tsout_ens(X, compr=2, fast=FALSE)
}