| kedd-package | R Documentation |
Kernel smoothing of a univariate density f and of its r-th
derivative f^{(r)}:
\hat{f}^{(r)}_h(x) = \frac{1}{n h^{r+1}} \sum_{i=1}^{n}
K^{(r)}\!\left(\frac{x - X_i}{h}\right),
with S4 classes, ggplot2 graphics and Rcpp/OpenMP data-driven bandwidth selectors.
Kernels. Every function accepts kernel as one of
"gaussian", "epanechnikov", "uniform",
"triangular", "biweight", "triweight",
"tricube", "cosine", "silverman", "logistic",
or a member "pow(s,t)" of the power family
K(x)=c_{s,t}(1-|x|^s)^t\mathbf{1}_{|x|\le 1} (see
kernel.pow).
Maximum derivative order r for each kernel (see the section
below).
Bandwidth selectors (S4 class bandwidth):
h.amise | asymptotic MISE (normal reference) |
h.ucv | unbiased (least-squares) cross-validation |
h.bcv | biased cross-validation (BCV1, BCV2) |
h.ccv | complete cross-validation |
h.mcv | modified cross-validation |
h.tcv | trimmed cross-validation |
h.scv | smoothed cross-validation (Gaussian) |
h.mlcv | maximum-likelihood cross-validation |
Estimation and kernels: dkde (S4 class
dkde), kernel.fun (K^{(r)}, class kfun),
kernel.conv (K^{(r)} * K^{(r)}, class kconv).
All classes have show() and ggplot2-based plot() methods;
lines() on a bandwidth object returns a layer to add with
+.
Parallelism: the O(n^2) criteria run in C++ with OpenMP. The
nthreads argument sets the threads of each call: 0 (default) =
all processors, k > 0 = k threads, capped by
OMP_THREAD_LIMIT; it does not change results nor later calls.
Cpp_omp_threads(nthreads) reports the threads actually used.
Data: five samples of size 200 from Marron and Wand (1992) normal
mixtures, kurtotic, outlier, bimodal, trimodal
and claw (see kedd-data).
Source code and bug reports: https://github.com/ac-guidoum/kedd and https://github.com/ac-guidoum/kedd/issues.
Documentation: a worked introduction is given in
vignette("kedd", package = "kedd"), and demonstrations in
demo("kedd"), demo("kernels") and demo("bandwidths").
Highest derivative order r (deriv.order) for each kernel.
Column 2: K^{(r)} is not identically zero (kernel.fun,
kernel.conv, dkde); above it these functions
return zero. Columns 3 to 5: the selector criterion is finite; above it the
selectors return h = NA with a warning. "any" means no limit and "–"
means not available, even at r = 0.
| kernel | kernel.fun, kernel.conv, dkde | UCV, BCV, TCV, AMISE | MCV | CCV |
gaussian | any | any | any | any |
cosine | any | 1 | 0 | -- |
silverman | any | any | any | any |
logistic | 20 | 8 | 8 | 8 |
epanechnikov | 2 | 1 | 0 | -- |
uniform | 0 | 0 | -- | -- |
triangular | 1 | 0 | -- | -- |
biweight | 4 | 2 | 1 | 0 |
triweight | 6 | 3 | 2 | 1 |
tricube | 9 | 3 | 2 | 1 |
pow(s,t) | \min(st, 16) | \min(r^*, 6) | \min(r^*-1, 6) | \min(r^*-2, 6) |
For pow(s,t), st applies when s and t are integers
(16 otherwise), and r^* is the largest integer
r < t + 1/2, with 2r \le st when s and t
are integers and r < s + 1/2 when s is not an even integer.
For example pow(2,3): 6, 3, 2, 1; pow(2,8): 16, 6, 6, 6;
pow(1.5,2.5): 16, 1, 0, –.
h.scv (Gaussian kernel) has no limit and h.mlcv
is for r = 0 only.
Arsalane Chouaib Guidoum acguidoum@univ-tam.dz (author, maintainer), Amel Henider amel.henider@univ-tam.dz (author), Iago Giné-Vázquez iago.gin-vaz@protonmail.com (contributor).
Guidoum, A. C. (2020). Kernel estimator and bandwidth selection for density and its derivatives: the kedd package. arXiv:2012.06102.
Marron, J. S. and Wand, M. P. (1992). Exact mean integrated squared error. The Annals of Statistics, 20, 712–736.
Wand, M. P. and Jones, M. C. (1995). Kernel Smoothing. Chapman and Hall, London.
dkde, kernel.fun, kernel.conv,
kernel.pow, h.ucv, h.scv,
bandwidth, kedd-data,
vignette("kedd"), citation("kedd").
x <- rnorm(150)
bw <- h.ucv(x, kernel = "biweight")
bw
plot(dkde(data = x, h = bw@h, kernel = "biweight"))
## logistic and power kernels
h.bcv(x, whichbcv = 2, kernel = "logistic")
h.bcv(x, whichbcv = 2, kernel = kernel.pow(2, 4))
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.