AML chimerism dataset - tuning ranger with mlr3

Libraries

set.seed(42)
library(tidyverse)
library(mlr3verse)
library(mlr3tuning)
library(mlr3tuningspaces)

Data

Read in data

dat = read.csv("./data/aml.all.df.csv")

Make outcome a binary variable (0/1 relapse)

dat$rbin = factor(dat$rbin, levels = c("yes", "no"))

Filter out any tests that are post-relapse

dat = dat[which(dat$bdate < dat$dor | is.na(dat$dor)), ]

Filter out relapse >720 days

dat = dat[which(dat$rbin == "no" | dat$rtime < 720),]

Filter out any missing tests

dat = dat[!is.na(dat$bmc_cdw) & !is.na(dat$bmc_cd3) & 
                  !is.na(dat$bmc_cd15) & !is.na(dat$bmc_cd34) &
                  !is.na(dat$pbc_cdw) & !is.na(dat$pbc_cd3) & 
                  !is.na(dat$pbc_cd15) & !is.na(dat$pbc_cd34),]

Get \(p(relapse)\) for baseline model

prbin = sum(as.numeric(dat$rbin)-1) / nrow(dat)

# dat2 <- dat %>%
#   select(rbin, txage, hla, tbi, abd, ci, mtx, mmf, agvhd, cgvhd,
#          bmc_cdw, bmc_cd3, bmc_cd15, bmc_cd34, 
#          pbc_cdw, pbc_cd3, pbc_cd15, pbc_cd34, ID)
dat2 <- dat %>%
  select(rbin, sex, txage, 
         rstatprtx, ghgp, tbi, 
         bmc_cdw, bmc_cd3, bmc_cd15, bmc_cd34, 
         pbc_cdw, pbc_cd3, pbc_cd15, pbc_cd34, ID)

dat2 <- dat2 %>% 
  mutate_if(is.character, as.factor)  %>% 
  mutate_if(is.integer, as.numeric) %>%
  # mutate(abd = tolower(abd)) %>%
  drop_na() %>%
  droplevels()

Tuning

Set up task

task_chim <- TaskClassif$new(id = "all", backend = dat2, 
                              target = "rbin")

Define patients for use in cross validation

# task_chim$col_roles$group <- "ID"
task_chim$set_col_roles("ID", remove_from = 'feature')

print keys and learners

as.data.table(mlr_tuning_spaces)

##                         key                                 label
##  1:  classif.glmnet.default       Classification GLM with Default
##  2:     classif.glmnet.rbv2     Classification GLM with RandomBot
##  3:    classif.kknn.default      Classification KKNN with Default
##  4:       classif.kknn.rbv2    Classification KKNN with RandomBot
##  5:  classif.ranger.default    Classification Ranger with Default
##  6:     classif.ranger.rbv2  Classification Ranger with RandomBot
##  7:   classif.rpart.default     Classification Rpart with Default
##  8:      classif.rpart.rbv2   Classification Rpart with RandomBot
##  9:     classif.svm.default       Classification SVM with Default
## 10:        classif.svm.rbv2     Classification SVM with RandomBot
## 11: classif.xgboost.default   Classification XGBoost with Default
## 12:    classif.xgboost.rbv2 Classification XGBoost with RandomBot
## 13:     regr.glmnet.default           Regression GLM with Default
## 14:        regr.glmnet.rbv2         Regression GLM with RandomBot
## 15:       regr.kknn.default          Regression KKNN with Default
## 16:          regr.kknn.rbv2        Regression KKNN with RandomBot
## 17:     regr.ranger.default        Regression Ranger with Default
## 18:        regr.ranger.rbv2      Regression Ranger with RandomBot
## 19:      regr.rpart.default         Regression Rpart with Default
## 20:         regr.rpart.rbv2       Regression Rpart with RandomBot
## 21:        regr.svm.default           Regression SVM with Default
## 22:           regr.svm.rbv2         Regression SVM with RandomBot
## 23:    regr.xgboost.default       Regression XGBoost with Default
## 24:       regr.xgboost.rbv2     Regression XGBoost with RandomBot
##                         key                                 label
##             learner n_values
##  1:  classif.glmnet        2
##  2:  classif.glmnet        2
##  3:    classif.kknn        3
##  4:    classif.kknn        1
##  5:  classif.ranger        4
##  6:  classif.ranger        8
##  7:   classif.rpart        3
##  8:   classif.rpart        4
##  9:     classif.svm        4
## 10:     classif.svm        5
## 11: classif.xgboost        8
## 12: classif.xgboost       13
## 13:     regr.glmnet        2
## 14:     regr.glmnet        2
## 15:       regr.kknn        3
## 16:       regr.kknn        1
## 17:     regr.ranger        4
## 18:     regr.ranger        7
## 19:      regr.rpart        3
## 20:      regr.rpart        4
## 21:        regr.svm        4
## 22:        regr.svm        5
## 23:    regr.xgboost        8
## 24:    regr.xgboost       13
##             learner n_values

tune learner with default search space

instance_rf = tune(
  method = "grid_search",
  task = task_chim,
  learner = lts(lrn("classif.ranger")),
  resampling = rsmp ("cv", folds = 5),
  measure = msr("classif.bacc"),
  term_evals = 100
)

instance_rf$result

##    mtry.ratio replace sample.fraction num.trees learner_param_vals  x_domain
## 1:          1    TRUE               1      1556          <list[5]> <list[4]>
##    classif.bacc
## 1:     0.754902

---

1. Stanford Medicine, dcshyr@stanford.edu↩︎

2. University of Utah, simon.brewer@geog.utah.edu↩︎