ELMSurv (version 0.6)

ELMBJ: ELMBJ

Description

A Kernel Extreme Learning Machine Using the Buckley-James estimator

Usage

ELMBJ(x, y, Regularization_coefficient, kerneltype = 2,
  Kernel_para = c(2, 1))

Arguments

x

The covariates(predictor variables) of training data.

y

Survival time and censored status of training data. Must be a Surv survival object

Regularization_coefficient

Ridge or Tikhonov regularization parameter. Default value for ELMBJEN is 10000. It need be set by the user here when using a single base ELM survival model. Also known as \(C\) in the ELM paper.

kerneltype

Type of kernel matrix. kerneltype=1,a RBF kernel;kerneltype=2 , a linear kernel;kerneltype=3 ,a polynomial kernel;kerneltype=4, a sigmoid kernel.

Kernel_para

Parameters for different types of kernels. A single value for kerneltype=1 or 2. A vector for kerneltype=3 or 4.

Value

List of returned values

trainMSE	Mean Square Error(MSE) on training data.
newy	Esitmated survival times of training data by the Buckley-James estimator.
outputWeight	Weights of the output layer in ELM.

References

• Hong Wang et al (2018). A Survival Ensemble of Extreme Learning Machine. Applied Intelligence, DOI:10.1007/s10489-017-1063-4.

See Also

ELMBJEN

Examples

# NOT RUN {
set.seed(123)
require(ELMSurv)
require(survival)
#Lung DATA
data(lung)
lung=na.omit(lung)
lung[,3]=lung[,3]-1
n=dim(lung)[1]
L=sample(1:n,ceiling(n*0.5))
trset<-lung[L,]
teset<-lung[-L,]
rii=c(2,3)
#A kernel ELM base model
kerelmsurv=ELMBJ(trset[,-rii],Surv(trset[,rii[1]],trset[,rii[2]]))
#The traing MSE
tr_mse=kerelmsurv$trainMSE
#New survival times imputed for training data
y_impute=kerelmsurv$newy
# }