CAFE (version 1.8.0)

discontSmooth: A discontinuous smoother

Description

Calculates discontinuous smoother

Usage

discontSmooth(y,gamma)

Arguments

y

input vector

gamma

The gamma level can be roughly compared to the sliding window size in a normal continuous smoother. The gamma level determines how strict the algorithm functions; a higher level will correspond to fewer jumps. This cannot be larger than length(y). Must be a positive integer.

Value

Vector with same length as input y

Details

Uses the potts filter algorithm described by Friedrich et al.

References

Friedrich, F., Kempe, a, Liebscher, V., & Winkler, G. (2008). Complexity Penalized M-Estimation. Journal of Computational and Graphical Statistics, 17(1), 201-224. doi:10.1198/106186008X285591

Examples

#generate piecewise vector with gaussian noise
y <- 1:450
y[1:150] <- 2
y[151:300] <- 3
y[301:450] <- 1
y <- y + rnorm(450)

#calculate smoother
y_smooth <- discontSmooth(y,20)