Object Reference : Object View and Procedure Reference : Series
Perform wavelet outlier detection for the series.
Series View: series_name.waveoutlier(options)
Basic Options
threshtype=arg (default = “soft”)
Wavelet threshold type: “hard” (hard thresholding), “soft” (soft thresholding).
threshlim=arg (default = “universal”)
Wavelet threshold limit type: “universal” (universal), “adaptive” (universal adaptive), “minimax” (minimax), “sureshrink” (SureShrink), “fdr” (false discovery rate).
If “threshlim=sureshrink”, the grid length may be specified using “ssglen=”.
If “threshlim=fdr”, the significance level may be specified using “fdrsig=”
mad=arg (default = “gauss”)
Mean/median absolute deviation: “mean” (mean absolute deviation), “gauss” (median absolute deviation with Gaussian adjustment), “median” (median absolute deviation), “meanmedian” (mean median absolute deviation).
sslen=arg (default = 10)
Grid length used in determining the SureShrink limit.
fdrsig=arg (default = .05)
Significance level as a number between 0 and 1 for false discovery rate limit determination.
Force the dialog to appear from within a program.
Print results.
Wavelet Transform Options
transform=arg (default=“dwt”)
Wavelet transform type: “dwt” (discrete wavelet transform – DWT), “modwt” (maximum overlap DWT – MODWT). Note that when performing DWT, if the series length is not dyadic, a dyadic fix may be set with the “fixlen=” option
fixlen=arg (default=“mean”)
Fix dyadic lengths in DWT: “zeros” (pad remainder with zeros), “mean” (pad remainder with mean of series), “median” (pad remainder with median of series), “shorten” (cut series length to dyadic length preceding series length).
maxscale=integer (default = max possible)
Maximum scale for wavelet transform.
The max possible is obtained as follows. Let denote the series length and decompose into its dyadic component and a remainder: , . The default maxscale is then set with the following rules:
DWT: (1) if then , otherwise (2) if expanding the series, and (3) if contracting the series .
filter=arg (default=“h”)
Wavelet filter class: “h” (Haar), “d” (Daubechies), “la” (least asymmetric).
If “filter=h” or “filter=la”, the filter length may be specified using “flen=”.
Wavelet filter boundary conditions are specified using the “bound=” option
Wavelet filter excess length as an even number between 2 and 20.
For use when “filter=d” (default= 4) or “filter=la” (default=8).
bound=arg (default = “p”)
Filter boundary handling: “p” (periodic), “r” (reflective).
dgp.wavetoutlier(maxscale=1, threshtype=hard, threshlim=meanmedian)
The line above will perform the Bilen and Huzurbazar (2002) wavelet outlier detection procedure on a series called DGP. It will use the Haar wavelet filter by default, execute to single scale, will use hard thresholding, and the mean median absolute deviation for the threshold limit. The latter options are those used in the original paper.
See “Wavelet Analysis” and “Wavelet Variance Decomposition” for discussion. See also “Wavelet Objects”.
See also Series::wavedecomp, Series::waveoutlier, Series::wavethresh, and Series::makewavelets.