cvscore: Calculate the CV score for determining regularisation...

In DeconWK: Deconvolution by Weighted Kernels

Description

Evaluates the cross-validation criterion (11) of Hazelton and Turlach (2009).

Usage

cv.score(y, sigma, h, gamma,
         METHOD=c("exact", "svm"), K=5, verb=FALSE)

Arguments

y	the observed values.
sigma	the standard deviation of the contaminating (normal) distribution.
h	the smoothing parameter to be used.
gamma	vector of values from which a suitable value is to be selected
METHOD	method to be used to solve the quadratic programming problem involved in calculating the weights; if "exact" then solveqp is used, otherwise the routine ipop from the kernlab package is used.
K	number of folds to be used if gamma is chosen by cross-validation; defaults to 5.
verb	logical; if TRUE some progress report will be printed during cross-validation.

Value

A vector containing the cross-validation criterion evaluated at the values given in gamma.

Author(s)

Berwin A Turlach [e-mail]

References

Hazelton, M.L. and Turlach, B.A. (2009). Nonparametric density deconvolution by weighted kernel estimators, Statistics and Computing 19(3): 217–228. http://dx.doi.org/10.1007/s11222-008-9086-7.

See Also

w.hat

Examples

  set.seed(100719)
  sig <- sqrt(29/40)  # Var(Z)/Var(X) = 0.1
  y <- rden(100, DEN=3, sigma=sig)
  h <- bw.SJ(y, method="dpi")
  gamma.ridge <- exp(seq(from=0, to=6, length=17))

  save.seed <- .Random.seed
  cv1 <- cv.score(y, sigma=sig, h=h, gamma=gamma.ridge,
                  METHOD="exact", verb=TRUE)
  plot(log(gamma.ridge), cv1, type="b")
  tmp <- getmin(log(gamma.ridge), cv1, which="r")
  abline(v=tmp$xmin)

  .Random.seed <- save.seed
  cv2 <- cv.score(y, sigma=sig, h=h, gamma=gamma.ridge,
                  METHOD="svm", verb=TRUE)
  plot(log(gamma.ridge), cv2, type="b")
  tmp <- getmin(log(gamma.ridge), cv2, which="r")
  abline(v=tmp$xmin)

  .Random.seed <- save.seed
  cv1 <- cv.score(y, sigma=sig, h=h, gamma=gamma.ridge,
                  METHOD="exact", K=10, verb=TRUE)
  plot(log(gamma.ridge), cv1, type="b")
  tmp <- getmin(log(gamma.ridge), cv1, which="r")
  abline(v=tmp$xmin)

  .Random.seed <- save.seed
  cv2 <- cv.score(y, sigma=sig, h=h, gamma=gamma.ridge,
                  METHOD="svm", K=10, verb=TRUE)
  plot(log(gamma.ridge), cv2, type="b")
  tmp <- getmin(log(gamma.ridge), cv2, which="r")
  abline(v=tmp$xmin)

DeconWK documentation built on May 2, 2019, 6:08 p.m.