README.md

filtrs

Rust-Boosted Linear and Spatial Filtering in R.

Currently, the package supports only the Whittaker-Eilers smoother as it is implemented in the whittaker-eilers crate. Based on that filter, a smoothing approach for spatial geometries (only single-part LINESTRING for now) is proposed.

Installation

You can install the development version of filtrs like so:

# install.packages(remotes)
remotes::install_github("atsyplenkov/filtrs")

Time-series filtering

For equally-spaced data, one can use the fil_wt function as an API to whittaker_eilers::WhittakerSmoother. It has two controls, lambda and order, and interpolates missing values by default. An extremely nice description of the smoother is written by Andrew Bowell.

library(filtrs)
## basic example code

data("airquality")

airquality$Ozone_smooth <- 
 fil_wt(as.double(airquality$Ozone), 10, 2)

plot(airquality$Ozone, type = "l", xlab = "Days", ylab = "Ozone")
lines(airquality$Ozone_smooth, col = "red", lwd = 2)

Spatial filtering

Spatial data may not be equally spaced; it often has more nodes on the bends of lines and fewer on straight segments. To address this challenge, the distance between nodes—calculated using either Cartesian or Haversine methods, depending on the Coordinate Reference System (CRS)—is used as a positions vector in the background. The function fil_wt_sf is designed to smooth a single LINESTRING, proving particularly beneficial for processing high-frequency data, such as that obtained from satellite imagery.

library(sf)
#> Linking to GEOS 3.11.2, GDAL 3.8.2, PROJ 9.3.1; sf_use_s2() is TRUE
library(smoothr)
#> 
#> Attaching package: 'smoothr'
#> The following object is masked from 'package:stats':
#> 
#>     smooth

file_path <- system.file("exdata/examples.gpkg", package = "filtrs")

lines <- 
  sf::st_read(file_path, layer = "gswe", quiet = TRUE) |>
  # Increase nodes count for smoother result
  smoothr::smooth("densify", max_distance = 10)

lines_wt <- 
  fil_wt_sf(lines, lamda = 10^-7, order = 3)

lines_smoothr <-
  smoothr::smooth(lines, method = "ksmooth",
                  smoothness = 21)

Plot’s code

par(
  mar = c(0.5, 0.5, 0.2, 0.2),
  mfrow = c(1, 2),
  oma = c(0, 0, 0.2, 0.2)
)

plot(
  sf::st_geometry(lines),
  col = "grey30",
  lwd = 3.5
)
plot(
  sf::st_geometry(lines_wt),
  col = 'firebrick3',
  lwd = 2,
  add = TRUE
)

# Add the legend
legend(
  "bottomleft",
  legend = c("Original", "{filtrs}", "{smoothr}"),
  col = c("grey30", "firebrick3", "dodgerblue3"),
  lwd = c(3, 3)
)

plot(
  sf::st_geometry(lines),
  col = "grey30",
  lwd = 3.5
)
plot(
  sf::st_geometry(lines_smoothr),
  col = 'dodgerblue3',
  lwd = 2,
  add = TRUE
)

Similar filtering can be achieved via the smoothr R package; however, the Rust-based approach outperforms in speed when high-order smoothing is needed. For example, 182 nodes were filtered 30 times faster using the filtrs package rather than smoothr.

bench::mark(
  filtrs = fil_wt_sf(lines, lamda = 10^-7, order = 3),
  smoothr = smoothr::smooth(lines, method = "ksmooth", smoothness = 21),
  check = F,
  relative = T
)
#> # A tibble: 2 × 6
#>   expression   min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <dbl>  <dbl>     <dbl>     <dbl>    <dbl>
#> 1 filtrs       1      1        32.4      1        2.95
#> 2 smoothr     35.6   34.1       1        8.19     1

Similar packages

As linear and spatial filters are pretty common, there is no shortage of analogs.

• smoothr - Spatial feature smoothing written in base R. However, there is no Whittaker-Eilers smoother.

• phenofit - As part of vegetation phenology, Whittaker-Eilers and Savitzky-Golay time-series filtering is implemented using Rcpp. No spatial filtering.

• signal - A set of signal processing functions originally written for ‘Matlab’ and ‘Octave’. No spatial filtering.