Internal Tuning

[sebffischer]

• [x] check that tuning works with branches in pipelines (where each branch contains an InternalTuneToken)
• [x] implement and call assert_internal_tuning() before starting the tuning process
• [x] check that AT works with instantiated resampling (this does not matter)
• [x] Implement it for the async variant
• [x] implement it for multi-grid
• [x] no validate field for the auto-tuner for now (?) (this is technically necessary)
• [x] implement set_validate() properly
• [x] more tests
• [x] Make it work with empty primary search space but existing internal search space (--> TunerInternal)

library(mlr3tuning)
library(mlr3learners)
library(mlr3pipelines)

# graph = ppl("robustify") %>>%
graph = ppl("robustify") %>>% lrn("classif.xgboost", alpha = to_tune(0, 1), nrounds = to_tune(upper = 1000L, internal = TRUE), early_stopping_rounds = 10L)

learner = as_learner(graph)

set_validate(learner, validate = "test")

at = auto_tuner(
  tuner = tnr("random_search"),
  learner = learner,
  resampling = rsmp("cv", folds = 3),
  measure = msr("classif.ce"),
  term_evals = 4
)

task = tsk("iris")

rr = resample(task, at, rsmp("cv", folds = 3L))