Learner has not been trained yet

[MislavSag]

I have trained models using mlr3batchmark package:

  # Exp dir
  date_ = strftime(Sys.time(), format = "%Y%m%d")
  dirname_ = glue("experiments_pread_live_{date_}")
  if (dir.exists(dirname_)) system(paste0("rm -r ", dirname_))

  # Create registry
  packages = c("data.table", "gausscov", "paradox", "mlr3", "mlr3pipelines",
               "mlr3tuning", "mlr3misc", "future", "future.apply",
               "mlr3extralearners", "stats")
  reg = makeExperimentRegistry(file.dir = dirname_, seed = 1, packages = packages)
  
  # Populate registry with problems and algorithms to form the jobs
  batchmark(designs, store_models = TRUE, reg = reg)
  
  # Save registry
  saveRegistry(reg = reg)

  # get nondone jobs
  ids = findNotDone(reg = reg)
  
  # set up cluster (for local it is parallel)
  cf = makeClusterFunctionsSocket(ncpus = 4L)
  reg$cluster.functions = cf
  saveRegistry(reg = reg)
  
  # define resources and submit jobs
  resources = list(ncpus = 2, memory = 8000)
  submitJobs(ids = ids$job.id, resources = resources, reg = reg)

When I load models after the training:

  results = reduceResultsBatchmark(reg = reg, store_backends = TRUE)
  
  # predictions for new dataset
  task_new = task$clone()
  task_new$filter(rows = data_ids)
  results$learners$learner[[1]]$predict(task = task_new)

I get an error

Error: Cannot predict, Learner 'filter_target_branch.nop_filter_target.filter_target_id.filter_target_unbranch.subsample.dropnacol.dropna.removeconstants_1.fixfactors.winsorizesimple.removeconstants_2.dropcorr.scale_branch.uniformization.scale.scale_unbranch.dropna_v2.nop_union_pca.pca.ica.feature_union_pca.disr.jmim.jmi.mim.mrmr.njmim.cmim.carscore.information_gain.relief.gausscov_f1st.feature_union_filters.removeconstants_3.regr.ranger.tuned' has not been trained yet

It returns an error like the model is not saved, even if I set I want to save it.