kernelTDA (version 1.0.0)

krein.svm.default: Krein Support Vector Machine

Description

Solves a kernelized Support Vector Machine in the case where the kernel used may not be positive semidefinite.

Usage

# S3 method for default
krein.svm(
  kernelmat = NULL,
  y = NULL,
  cost = 1,
  class.weights = NULL,
  cross = 0,
  probability = FALSE,
  fitted = TRUE,
  subset,
  ...
)

Arguments

kernelmat

the kernel matrix computed for all observations

y

a vector of labels

cost

cost of violating the constraint

class.weights

a named vector of weights for the different classes, used for asymmetric class sizes. Not all factor levels have to be supplied (default weight: 1). All components have to be named. Specifying "inverse" will choose the weights inversely proportional to the class distribution.

cross

number of fold in a k-fold cross validation

probability

logical indicating whether the model should allow for probability predictions (default: FALSE).

fitted

logical indicating whether the fitted values should be computed and included in the model or not (default: TRUE)

subset

an index vector specifying the cases to be used in the training sample. (NOTE: If given, this argument must be named.)

...

additional parameters

Value

An object of class krein.svm containing the fitted model, including:

SV

a matrix containing the Support Vectors

index

index of the resulting support vectors in the data matrix

coefs

a matrix containing corresponding coefficients times the training labels

rho

value of the (negative) intercept

Details

This function implements the Krein Support Vector Machine solver as defined by Loosli et al. (2015). The implementation of the solver is a modified version of the popular C++ library `LIBSVM`, while the connection to `R` heavily relies on the `R`-package e1701.

References

loosli2015learningkernelTDA

chang2011libsvmkernelTDA

dimitriadou2008misckernelTDA

Examples

# NOT RUN {
## DO NOT RUN:
# library(TDA)
# set.seed(123)
# foo.data = list()
# for(i in 1:20){
#    foo = circleUnif(100)
#    foo.data[[i]] = ripsDiag(foo, 1,1)$diagram}
#    for(i in 21:40){   
#     foo = cbind(runif(100), runif(100))
#     foo.data[[i]] = ripsDiag(foo, 1,1)$diagram
#     }
# GSWkernel = gaus.kernel(foo.data, h =1, dimension = 1,  q = 2)
# GGKclass = krein.svm(kernelmat = GSWkernel, y = rep(c(1,2), c(20,20)))
# }