Fit weighted kernel svm model.

Fits weighted kernel SVM. To be used for OWL with hinge loss (but can be used more generally)

weighted.ksvm(
  y,
  x,
  weights,
  C = c(0.1, 0.5, 1, 2, 10),
  kernel = "rbfdot",
  kpar = "automatic",
  nfolds = 10,
  foldid = NULL,
  eps = 1e-08,
  ...
)

Arguments

y: The response vector (either a character vector, factor vector, or numeric vector with values in -1, 1)
x: The design matrix (not including intercept term)
weights: vector of sample weights for weighted SVM
C: cost of constraints violation, see ksvm
kernel: kernel function used for training and prediction. See ksvm and kernels
kpar: list of hyperparameters for the kernel function. See ksvm
nfolds: number of cross validation folds for selecting value of C
foldid: optional vector of values between 1 and nfolds specifying which fold each observation is in. If specified, it will override the nfolds argument.
eps: penalty nugget parameter. Defaults to 1e-8
...: extra arguments to be passed to ipop from the kernlab package

Examples


library(kernlab)
#> 
#> Attaching package: ‘kernlab’
#> The following object is masked from ‘package:ggplot2’:
#> 
#>     alpha

x <- matrix(rnorm(200 * 2), ncol = 2)

y <- 2 * (sin(x[,2]) ^ 2 * exp(-x[,2]) - 0.2 > rnorm(200, sd = 0.1)) - 1

weights <- runif(100, max = 1.5, min = 0.5)

wk <- weighted.ksvm(x = x[1:100,], y = y[1:100],
                    C = c(0.1, 0.5, 1, 2),
                    nfolds = 5,
                    weights = weights[1:100])

pr <- predict(wk, newx = x[101:200,])

mean(pr == y[101:200])
#> [1] 0.78

Arguments

See also

Examples