How to perform stacking of multiple survival learners

[TianshuGu]

library(data.table)
library(mlr3)
library(mlr3proba)
library(mlr3pipelines)
library(mlr3learners)
library(mlr3extralearners)
library(mlr3tuning)
library(mlr3benchmark)
library(glmnet)
library(ggplot2)
library(survex)
library(tidyr)
library(survival)
# remotes::install_github("mlr-org/mlr3filters")
library(mlr3filters)

data("lung", package = "survival")
data <- as.data.table(lung)

# 
data$time <- data$time / 30  # 将生存时间从天转换为月（可选）

#  
data <- data[, lapply(.SD, function(x) ifelse(is.na(x), median(x, na.rm = TRUE), x))]

 

#  
task <- TaskSurv$new(id = "lung_survival_task", backend = data,time = 'time',event = 'event')


# （Base Learners）
base_learners = lrns(c("surv.coxph", "surv.kaplan", "surv.cv_glmnet", "surv.xgboost"))

# （Super Learner）
super_learner = lrn("surv.ranger" )

# stacking
stacking = ppl("stacking",
               base_learners = base_learners,
               super_learner = super_learner
)

# training
stacking$train(lung_task)


###This is my error message

> stacking$train(lung_task)
INFO  [16:33:19.810] [mlr3] Applying learner 'surv.coxph' on task 'lung_survival_task' (iter 1/3)
INFO  [16:33:19.833] [mlr3] Applying learner 'surv.coxph' on task 'lung_survival_task' (iter 2/3)
INFO  [16:33:19.854] [mlr3] Applying learner 'surv.coxph' on task 'lung_survival_task' (iter 3/3)
INFO  [16:33:19.935] [mlr3] Applying learner 'surv.kaplan' on task 'lung_survival_task' (iter 1/3)
INFO  [16:33:19.946] [mlr3] Applying learner 'surv.kaplan' on task 'lung_survival_task' (iter 2/3)
INFO  [16:33:19.956] [mlr3] Applying learner 'surv.kaplan' on task 'lung_survival_task' (iter 3/3)
INFO  [16:33:20.241] [mlr3] Applying learner 'surv.cv_glmnet' on task 'lung_survival_task' (iter 1/3)
INFO  [16:33:20.402] [mlr3] Applying learner 'surv.cv_glmnet' on task 'lung_survival_task' (iter 2/3)
INFO  [16:33:20.580] [mlr3] Applying learner 'surv.cv_glmnet' on task 'lung_survival_task' (iter 3/3)
INFO  [16:33:20.839] [mlr3] Applying learner 'surv.xgboost' on task 'lung_survival_task' (iter 1/3)
INFO  [16:33:20.856] [mlr3] Applying learner 'surv.xgboost' on task 'lung_survival_task' (iter 2/3)
INFO  [16:33:20.871] [mlr3] Applying learner 'surv.xgboost' on task 'lung_survival_task' (iter 3/3)
Error: <TaskSurv:lung_survival_task> has the following unsupported feature types: list
This happened PipeOp surv.ranger's $train()
In addition: Warning messages:
1: In coxph.fit(X, Y, istrat, offset, init, control, weights = weights,  :
  Loglik converged before variable  8 ; coefficient may be infinite. 
This happened PipeOp surv.coxph.surv.coxph's $train()
2: In coxph.fit(X, Y, istrat, offset, init, control, weights = weights,  :
  Loglik converged before variable  8 ; coefficient may be infinite. 
This happened PipeOp surv.coxph.surv.coxph's $train()
3: In coxph.fit(X, Y, istrat, offset, init, control, weights = weights,  :
  Loglik converged before variable  8 ; coefficient may be infinite. 
This happened PipeOp surv.coxph.surv.coxph's $train()
4: In coxph.fit(X, Y, istrat, offset, init, control, weights = weights,  :
  Loglik converged before variable  8 ; coefficient may be infinite. 
This happened PipeOp surv.coxph.surv.coxph's $train()

This always happened on the super learner

[sebffischer]

@jemus42

[jemus42]

I have not done any stacking yet, but the book has a section on that so I hope that helps?