diss.DWT {TSclust}

Dissimilarity for Time Series Based on Wavelet Feature Extraction

Description

Performs an unsupervised feature extration using orthogonal wavelets on the series and returns the Euclidean distance between the wavelet approximations in an appropriate scale.

Usage

diss.DWT(series)

Arguments

series

Numeric matrix with row order time series

Details

This method differs from other dissimilarities in that pairwise dissimilaries depend on the whole dataset that is given to diss.DWT, hence, there is no pairwise version of the function defined, only accepts whole datasets. The set of original series is replaced by their wavelet approximation coefficients in an appropriate scale, and the dissimilarity between two series is computed as the Euclidean distance between these coefficients. The appropriate scale is automatically determined by using an algorithm addressed to obtain an efficient reduction of the dimensionality but preserving as much information from the original data as possible. The algorithm is introduced by Zhang, Ho, Zhang, and Lin (2006).

Value

Returns an object of type dist with the pairwise distances.

Author(s)

Pablo Montero Manso, José Antonio Vilar.

References

Zhang, H., Ho, T. B., Zhang, Y., and Lin, M. (2006) Unsupervised feature extraction for time series clustering using orthogonal wavelet transform. INFORMATICA-LJUBLJANA-, 30(3), 305.

Montero, P and Vilar, J.A. (2014) TSclust: An R Package for Time Series Clustering. Journal of Statistical Software, 62(1), 1-43. http://www.jstatsoft.org/v62/i01/.

Examples

## Create three sample time series
x <- cumsum(rnorm(100))
y <- cumsum(rnorm(100))
z <- sin(seq(0, pi, length.out=100))

#compute the distance
diss.DWT(rbind(x, y, z))