I am tuning parameters for multiple learners using parallel computing. However, I am encountering this error message for the learner ‘classif.svm‘:

<simpleError: '<OptimizerMbo>' does not support param sets with dependencies!>

I don’t understand why I am receiving this error message because there doesn’t seem to be any dependency according to this command:

> learner_svm$param_set$values
$type
[1] "C-classification"

$cost
Tuning over:
range [1e-04, 10000] (log scale)


$kernel
Tuning over:
p_fct(levels = c("polynomial", "radial", "sigmoid", "linear"))

$degree
Tuning over:
range [2, 5]


$gamma
Tuning over:
range [1e-04, 10000] (log scale)

Why am I encountering this issue? But then again, is this normal because in the book, it is noted that SVM models have dependencies?

Here is an example to reproduce this problem. I have ten different learners, but to speed things up, I’m only using two of them:

library(mlr3)
library(mlr3learners)
library(mlr3verse)
library(mlr3spatial)
library(mlr3spatiotempcv)
library(mlr3tuning)
library(mlr3pipelines)

data <- data.frame(ID = 1:1742, x = runif(1742, -130.88, -61.12), y = runif(1742, 12.12, 61.38), year = runif(1742, 2005, 2020), presence = rep(0:1, each=871), 
                      V1 = runif(1742, -3.66247, 2.95120), V2 = runif(1742, -1.6501, 7.5510))
## head(data)


data$presence <- as.factor(data$presence)
classification_task_sp <- mlr3spatial::as_task_classif_st(x = data, target = response, positive = "1", coordinate_names = c("x", "y"), crs = "+proj=longlat +datum=WGS84 +no_defs +type=crs")

classification_task_sp$set_col_roles(ID, roles = "space")
classification_task_sp$set_col_roles(year, roles = "time")

learner_glmnet <- mlr3tuningspaces::lts(mlr3::lrn("classif.glmnet", predict_type = "prob"))
learner_svm <- mlr3tuningspaces::lts(mlr3::lrn("classif.svm", predict_type = "prob", type  = "C-classification"))

learners <- c(learner_glmnet, learner_glmnet, learner_glmnet, learner_glmnet, learner_glmnet, learner_glmnet, learner_svm)

inner_resampling <- mlr3::rsmp("sptcv_cstf", folds = 2)
measures <- c(mlr3::msr("classif.sensitivity"), mlr3::msr("classif.specificity"), mlr3::msr("classif.acc"))
tuner <- mlr3tuning:::tnr("mbo")

cl <- parallel::makeCluster(4, outfile = "test.txt")
doSNOW::registerDoSNOW(cl)

tuned_hyperparameter_list <- foreach::foreach(learner_ID = 1:length(learners), .errorhandling = "pass", .packages = c("dplyr")) %dopar% {
  
  set.seed(1)
  tuning_instance <- mlr3tuning::ti(task = task,
                                           learner = learners[[learner_ID]],
                                           resampling = inner_resampling,
                                           measures = measures,
                                           terminator = mlr3tuning::trm("evals", n_evals = 1))
  
  ## Run the tuning process  
  tuner$optimize(tuning_instance)
  
  df <- data.frame(tuning_instance$result)
  ## print(df)
  
  ## Return a data frame
  return(df)
  
}

parallel::stopCluster(cl)