create_svhm: Train SVHM

In herglola/SVHM: Support-Vector-Hazard-Machine

Train SVHM

Description

predicts the event times of a given dataset using cross validation for the cost parameter. Final model includes predicted event times as well as parameters to predict new event times from subject who are not in df. All values of covariates are first normalized to the intervall [0,1] before the SVHM algorithm is applied. The cost parameter for the final model is chosen with the best pearson correlation.

Usage

create_svhm(
  df,
  covariates,
  cross_validation_val,
  cost_grid,
  varName_cencored,
  varName_futime,
  k = 3,
  test_size = 0.2,
  opt = "osqp",
  gamma_squared = 0.5,
  choose = "c"
)

Arguments

df	data frame
covariates	vector of name of covariates
cross_validation_val	number of subset to use for cost optimization
cost_grid	grid of all cost parameter to be optoimzed uponl
varName_cencored	name of variable in df that indicates cencoring
varName_futime	name of variable in df that indicates event time
k	integer of how many nearest event times are used to predict the event time (default is 3)
test_size	size of final test set in precent
opt	which quadratic optimization is used (opt='mosek' or opt='osqp')
gamma_squared	width of gaussian kernel
choose	optional parameter which decides if the C-index or the pearson correlation is used to determine the optimal cost parameter. Values are either 'c' for the C-Index or 'p' for the pearson correlation

Value

trained model with $e_vec vector indicating vector containing information if a subject is at risk or if an event happens. If n are the number of subjects and m the number of event times, then event_vec has length n*m, $k_mat kernel matrix, $sol calculated optimal solution, $t_predict test dataset with risk scores risk and t_predict, $p_corr pearson correlation of the predicted times $C_indes C-Index

Note

The mosek package requires a license

Examples

{

library(KMsurv)
library(SVHM)

##############
# Parameters #
##############

gamma_squared <- 100
k <- 1
cross_validation_val <- 3
test_size=.3
cost_grid <- 2^c(-6:6)

covariates <- c('z7')

######################
#  Model prediction  #
######################

data(bmt)

model <- create_svhm(bmt, covariates, cross_validation_val, cost_grid, varName_cencored="d3", varName_futime = "t2", k=k, test_size=test_size, opt='osqp', gamma_squared=gamma_squared)
}

herglola/SVHM documentation built on March 24, 2022, 12:44 p.m.