kedd — Kernel Estimator and Bandwidth Selection for Density and Its Derivatives kedd logo

CRAN version CRAN status R-CMD-check Total downloads Monthly downloads Weekly downloads Test coverage

The kedd package (version 2.0.2) provides nonparametric kernel smoothing tools for R: kernel estimation of a density and of its derivatives of any order r, kernel functions, their derivatives and convolutions, and nine data-driven bandwidth selectors. Results are S4 objects with ggplot2 graphics; the \(O(n^2)\) criteria run in C++ (Rcpp) with OpenMP.

Authors: Arsalane Chouaib Guidoum (author, maintainer) and Amel Henider (author), with Iago Giné-Vázquez (contributor) — see Authors.

Features

Kernel functions (10 named kernels + the Pow(s,t) family)

Kernel Formula Support r max
Gaussian \(K(x) = \frac{1}{\sqrt{2\pi}} e^{-x^2/2}\) \((-\infty, +\infty)\) any
Epanechnikov \(K(x) = \frac{3}{4}(1-x^2)\) \([-1, 1]\) 2
Uniform \(K(x) = \frac{1}{2}\) \([-1, 1]\) 0
Triangular \(K(x) = 1-\lvert x\rvert\) \([-1, 1]\) 1
Biweight \(K(x) = \frac{15}{16}(1-x^2)^2\) \([-1, 1]\) 4
Triweight \(K(x) = \frac{35}{32}(1-x^2)^3\) \([-1, 1]\) 6
Tricube \(K(x) = \frac{70}{81}(1-\lvert x\rvert^3)^3\) \([-1, 1]\) 9
Cosine \(K(x) = \frac{\pi}{4}\cos\left(\frac{\pi x}{2}\right)\) \([-1, 1]\) any
Silverman \(K(x) = \frac{1}{2}e^{-\lvert x\rvert/\sqrt{2}}\sin\left(\frac{\lvert x\rvert}{\sqrt{2}}+\frac{\pi}{4}\right)\) \((-\infty, +\infty)\) any
Logistic \(K(x) = \frac{e^{-x}}{(1+e^{-x})^2} = \frac{1}{4}\mathrm{sech}^2(x/2)\) \((-\infty, +\infty)\) 20
Pow(s,t) \(K(x) = c_{s,t}(1-\lvert x\rvert^s)^t\), \(c_{s,t} = \frac{s}{2B(1/s,\,t+1)}\) \([-1, 1]\) \(\min(st, 16)\)

Kernels with support \([-1, 1]\) are zero outside it. “r max” is the highest derivative order with \(K^{(r)} \ne 0\) (kernel.fun(), kernel.conv(), dkde()); for Pow(s,t) it is \(\min(st, 16)\) when \(s\) and \(t\) are integers and 16 otherwise. The selectors have lower limits, given in Maximum derivative order.

The power family is passed as kernel = "pow(s,t)" (or kernel.pow(s, t)) and contains uniform pow(s,0), triangular pow(1,1), Epanechnikov pow(2,1), biweight pow(2,2), triweight pow(2,3) and tricube pow(3,3). Any \(s > 0\), \(t \ge 0\) is allowed; derivatives up to order 16 are computed exactly and the constants \(\mu_2(K)\), \(\mu_4(K)\), \(R(K)\) come from Beta functions. Logistic derivatives are available up to order 20.

Bandwidth selectors

Function Method Derivative order r Reference
h.amise Asymptotic MISE rule (normal reference) 0, 1, 2, … (max: see below) Wand & Jones (1995)
h.ucv Unbiased (least-squares) Cross-Validation 0, 1, 2, … (max: see below) Rudemo (1982), Bowman (1984)
h.bcv Biased Cross-Validation (BCV1, BCV2) 0, 1, 2, … (max: see below) Scott & Terrell (1987)
h.ccv Complete Cross-Validation 0, 1, 2, … (max: see below) Jones & Kappenman (1992)
h.mcv Modified Cross-Validation 0, 1, 2, … (max: see below) Stute (1992)
h.tcv Trimmed Cross-Validation 0, 1, 2, … (max: see below) Feluch & Koronacki (1992)
h.scv Smoothed Cross-Validation (Gaussian kernel) 0, 1, 2, … (no limit) Hall, Marron & Park (1992)
h.pcv Partitioned Cross-Validation (UCV, TCV, MCV, CCV, BCV1 or BCV2 in m subsamples, nrep partitions) as the criterion Marron (1987)
h.mlcv Maximum-Likelihood Cross-Validation 0 only Habbema et al. (1974), Duin (1976)

With the Gaussian kernel every order r is allowed; with the other kernels r is limited by the regularity of the kernel, as given in the table Maximum derivative order r below.

All selectors return an S4 object of class bandwidth; the search interval is [0.01 * hos, 2 * hos] by default, hos being the over-smoothing bandwidth.

Maximum derivative order r

The regularity \(q\) of a kernel is the largest \(m\) such that the pointwise derivative \(K^{(m)}\) is the derivative of \(K\) in the sense of distributions (\(K^{(m-1)}\) absolutely continuous). For \(m > q\), \(K^{(m)}\) misses the Dirac masses at the jumps of \(K^{(q)}\), and a criterion that evaluates it at the pairwise differences no longer estimates its target. Each criterion therefore restricts \(r\): \(r \le q\) for dkde and h.amise, \(2r \le q\) for UCV and TCV (\(K^{(2r)}\)), \(2r + 2 \le q\) for MCV, \(r + 2 \le q\) for BCV1 (convolution of \(K^{(r+2)}\)) and \(2r + 4 \le q\) for BCV2 and CCV. \(K^{(r)}\) (\(K^{(r+2)}\) for BCV1) must also be square integrable, and the selectors are capped at \(r = 8\) (logistic) and \(r = 6\) (Pow), the derivative orders implemented being 20 and 16. Above the limit the selectors return h = NA with a warning, and dkde() gives a warning; “—” means not available even at \(r = 0\):

kernel \(q\) dkde h.amise h.ucv, h.tcv h.mcv BCV1 BCV2, h.ccv
gaussian \(\infty\) any any any any any any
logistic \(\infty\) 20 8 8 8 8 8
silverman 3 3 (*) 1 (*) (*) (*)
triweight, tricube 3 3 3 1 0 1 —
biweight 2 2 2 1 0 0 —
epanechnikov, triangular, cosine 1 1 1 0 — — —
uniform 0 0 0 0 — — —

() The Silverman kernel is of order 4 (\(\mu_2(K) = 0\)): h.amise() returns the value of the Gaussian kernel with a warning, and BCV1, BCV2, MCV and CCV, built on the bias expansion of a second-order kernel, are not defined for it. For pow(s,t), \(q = \lceil t + 1 \rceil - 1\) (\(q = t\) for an integer \(t\)) from \(|x| = 1\), and \(q = \lceil s + 1 \rceil - 1\) from \(x = 0\) unless \(s\) is an even integer; \(q\) is the smaller of the two. For example pow(2,4): \(q = 4\) and 4, 4, 2, 1, 2, 0; pow(1.5,2.5): \(q = 2\) and 1, 1, 1, 0, —, —. h.scv (Gaussian kernel) has no limit and h.mlcv is for \(r = 0\) only; h.pcv has the limits of its criterion. The same table is in ?kedd (section Maximum derivative order*).

Main functions

Installation

# Released version 2.0.2 on CRAN
install.packages("kedd")

# Development version from GitHub
# install.packages("remotes")
remotes::install_github("ac-guidoum/kedd", build_vignettes = TRUE)

# Version 2.0.2 from the source tarball
install.packages("kedd_2.0.2.tar.gz", repos = NULL, type = "source")

# or from the source directory
devtools::install("path/to/kedd_2.0.2")

Building from source compiles C++ code: on Windows this needs Rtools.

Quick start

library(kedd)
x <- rnorm(500)

# Kernel density estimate (bandwidth chosen by UCV when h is missing)
d <- dkde(data = x, kernel = "gaussian")
plot(d)

# First derivative with its own bandwidth
h1 <- h.scv(x, deriv.order = 1)
plot(dkde(data = x, h = h1@h, deriv.order = 1))

# Bandwidth selectors
h.ucv(x)       # unbiased CV
h.bcv(x)       # biased CV
h.scv(x)       # smoothed CV (Gaussian kernel)
h.pcv(x, m = 2)  # partitioned CV (UCV in 2 subsamples)

# Kernel function evaluation
kf <- kernel.fun(kernel = "epanechnikov", deriv.order = 1L)
plot(kf)

# Logistic and Pow(s,t) kernels
h.ucv(x, kernel = "logistic")
h.bcv(x, kernel = kernel.pow(2, 4))          # same as kernel = "pow(2,4)"
plot(kernel.fun(kernel = "pow(1.5,2.5)", deriv.order = 1L))

# Overlay two CV curves (ggplot2)
p <- plot(h.ucv(x))
p$plot + lines(h.bcv(x))

Datasets

Five samples of size 200 from the Marron & Wand (1992) normal mixtures, with known true densities (see ?"kedd-data"):

Dataset Marron & Wand density
kurtotic #4 kurtotic unimodal: \(\frac23 N(0,1) + \frac13 N(0, 0.1^2)\)
outlier #5 outlier: \(\frac1{10} N(0,1) + \frac9{10} N(0, 0.1^2)\)
bimodal #7 separated bimodal: \(\frac12 N(-1.5, 0.5^2) + \frac12 N(1.5, 0.5^2)\)
trimodal #9 trimodal: \(\frac9{20} N(-1.2, 0.6^2) + \frac9{20} N(1.2, 0.6^2) + \frac1{10} N(0, 0.25^2)\)
claw #10 claw: \(\frac12 N(0,1) + \sum_{l=0}^{4}\frac1{10} N(l/2-1, 0.1^2)\)
h.mcv(claw)
plot(dkde(data = bimodal, h = h.ucv(bimodal, deriv.order = 1)@h, deriv.order = 1))

Cheatsheet

A two-page summary of the package is available in cheatsheet/kedd-cheatsheet.pdf (source: cheatsheet/kedd-cheatsheet.html).

Vignette and demos

vignette("kedd", package = "kedd")   # theory, kernels, estimators, selectors, examples
demo(package = "kedd")               # list of demos
demo("kedd", package = "kedd")       # quick tour
demo("kernels", package = "kedd")    # kernels, derivatives, convolutions, Pow(s,t)
demo("bandwidths", package = "kedd") # all selectors compared on a bimodal sample

Tests

The testthat suite (≈1,850 expectations) covers every selector × kernel, the nthreads behaviour, the content of every ggplot2 figure (plot(), lines()), the datasets and input validation. It checks every kernel constant (\(\mu_2\), \(\mu_4\), \(R(K^{(r)})\)), convolution and derivative, for every kernel and every implemented order, against numerical integration, and checks Pow(s,t) against the named kernels it generalises (identical kernels, convolutions and bandwidths).

Line coverage measured with covr (tests only): 99.2 % overall.

File Coverage File Coverage
src/kedd.cpp 99.5 % R/h.mlcv.R 100 %
R/bandwidth-class.R 99.1 % R/h.pcv.R 95.7 %
R/dkde.R 97.8 % R/h.scv.R 100 %
R/h.amise.R 100 % R/h.tcv.R 100 %
R/h.bcv.R 100 % R/h.ucv.R 100 %
R/h.ccv.R 100 % R/kernel.fun.R 98.3 %
R/h.mcv.R 100 % R/utils.R 97.9 %
cv <- covr::package_coverage("path/to/kedd_2.0.2", type = "tests")
covr::percent_coverage(cv)                     # 99.2
covr::report(cv)                               # line-by-line HTML report
source("data-raw/benchmark.R")                 # execution time of every function

Downloads

CRAN downloads of kedd from the RStudio/Posit CRAN mirror, obtained with cranlogs (daily counts since the first release in August 2013) and cross-checked with dlstats (monthly counts; identical on the 70 months both cover, Jan 2021 – Oct 2026).

Monthly CRAN downloads of kedd
Statistic (as of 19 Nov 2026) Value
Total downloads (Aug 2013 – Nov 2026) 66,708
Last month 735
Last week 301
Monthly mean / median 417 / 298
Best month Sep 2025 (1,776)
Year 2013* 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026*
Downloads 586 2,472 3,872 2,818 2,918 3,856 7,657 12,813 10,566 4,844 743 3,462 6,260 3,841

* partial years (from Aug 2013; to 19 Nov 2026). The badges above are live; the figures in this section are a snapshot.

library(cranlogs)
sum(cran_downloads("kedd", from = "2013-08-01", to = Sys.Date() - 1)$count)  # total
cran_downloads("kedd", when = "last-month")

library(dlstats)
x <- cran_stats("kedd")          # monthly downloads
plot_cran_stats("kedd")

The figure and the numbers are regenerated with source("data-raw/downloads.R").

Authors

Name Role Contact
Arsalane Chouaib Guidoum author, maintainer · ORCID 0000-0003-3781-2160
Amel Henider author
Iago Giné-Vázquez contributor · ORCID 0000-0002-6725-2638

Source code: https://github.com/ac-guidoum/kedd · bug reports and suggestions: https://github.com/ac-guidoum/kedd/issues.

To cite the package (citation("kedd")):

Guidoum, A. C. (2020). Kernel estimator and bandwidth selection for density and its derivatives: the kedd package. arXiv:2012.06102. https://doi.org/10.48550/arXiv.2012.06102

References