Description
Efficient Savitzky-Golay Filtering.
Description
Smoothing signals and computing their derivatives is a common requirement in signal processing workflows. Savitzky-Golay filters are a established method able to do both (Savitzky and Golay, 1964 <doi:10.1021/ac60214a047>). This package implements one dimensional Savitzky-Golay filters that can be applied to vectors and matrices (either row-wise or column-wise). Vectorization and memory allocations have been profiled to reduce computational fingerprint. Short filter lengths are implemented in the direct space, while longer filters are implemented in frequency space, using a Fast Fourier Transform (FFT).
README.md
sgolay
The goal of sgolay is to offer efficient and vectorized Savitzky-Golay filters.
Installation
You can install the CRAN version with
# install.packages("sgolay")
Or you can install the development version of sgolay from GitHub with:
# install.packages("remotes")
remotes::install_github("zeehio/sgolay")
Benchmark
sgolay is faster than signal, especially on either larger filter lengths or when applied on matrices, since it uses the Fast Fourier Transform and avoids several memory copies and extra allocations.
library(sgolay)
x <- matrix(runif(1000*1000), nrow = 1000, ncol = 1000)
filt <- signal::sgolay(p = 2, n = 51)
timing <- bench::mark(
signal = apply(x, 2L, function(s) signal::sgolayfilt(s, filt)),
sgolay = sgolay::sgolayfilt(x, filt),
min_iterations = 50
)
#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
plot(timing, type = 'ridge')
#> Loading required namespace: tidyr
#> Picking joint bandwidth of 0.0135
