compatibility: work with new irace 4.0.0 (#435)

* compatibility: work with new irace 4.0.0

* fix: irace

* ...

* ...

DESCRIPTION

@@ -29,7 +29,7 @@ Depends:
     paradox (>= 1.0.1),
     R (>= 3.1.0)
 Imports:
-    bbotk (>= 1.4.0),
+    bbotk (>= 1.4.1),
     checkmate (>= 2.0.0),
     data.table,
     lgr,
@@ -39,7 +39,7 @@ Suggests:
     adagio,
     future,
     GenSA,
-    irace,
+    irace (>= 4.0.0),
     knitr,
     mlflow,
     mlr3learners (>= 0.7.0),

NEWS.md

@@ -13,6 +13,7 @@ perf: save models on worker only when requested in `ObjectiveTuningAsync`.
 * refactor: Rename stage `on_result` to `on_result_end` in `CallbackAsyncTuning` and `CallbackBatchTuning`.
 * docs: Extend the `CallbackAsyncTuning` and `CallbackBatchTuning` documentation.
 * compatibility: mlr3 0.22.0
+* compatibility: Work with new irace 4.0.0
 
 # mlr3tuning 1.1.0
 

R/TunerBatchIrace.R

@@ -132,14 +132,14 @@ TunerBatchIrace = R6Class("TunerBatchIrace",
   )
 )
 
-target_runner_tuning = function(experiment, exec.target.runner, scenario, target.runner) {# nolint
+target_runner_tuning = function(experiment, exec_target_runner, scenario, target_runner) {# nolint
   tuning_instance = scenario$targetRunnerData$inst
 
   xdt = map_dtr(experiment, function(e) {
     configuration = as.data.table(e$configuration)
     # add configuration and instance id to archive
-    set(configuration, j = "configuration", value = e$id.configuration)
-    set(configuration, j = "instance", value = e$id.instance)
+    set(configuration, j = "configuration", value = e$id_configuration)
+    set(configuration, j = "instance", value = e$id_instance)
     configuration
   })
   # fix logicals

tests/testthat/test_TunerBatchIrace.R

@@ -1,8 +1,7 @@
 skip_if_not_installed("irace")
-skip_on_cran()
 
 test_that("TunerIrace", {
-  x = capture.output({z = test_tuner("irace", term_evals = 42, real_evals = 39)})
+  z = test_tuner("irace", term_evals = 42)
   instance = z$inst
   archive = instance$archive$data
   tuner = z$tuner
@@ -12,7 +11,7 @@ test_that("TunerIrace", {
 
   # check optimization direction
   # first elite of the first race should have the lowest average performance
-  load(tuner$param_set$values$logFile)
+  iraceResults = irace::read_logfile(tuner$param_set$values$logFile)
   elites = iraceResults$allElites
   aggr = archive[race == 1, .(classif.ce = mean(classif.ce)), by = configuration]
   expect_equal(aggr[which.min(classif.ce), configuration], elites[[1]][1])
@@ -27,8 +26,13 @@ test_that("TunerIrace works with dependencies", {
     cp = p_dbl(lower = 0.001, upper = 0.1),
     minsplit = p_int(lower = 1, upper = 10, depends = cp == 0.005)
   )
-  instance = TuningInstanceBatchSingleCrit$new(tsk("iris"), lrn("classif.rpart"), rsmp("holdout"), msr("classif.ce"),
-    trm("evals", n_evals = 96), search_space)
+  instance = ti(
+    task = tsk("mtcars"),
+    learner = lrn("regr.rpart"),
+    resampling = rsmp("holdout"),
+    measures = msr("regr.mse"),
+    terminator = trm("evals", n_evals = 96),
+    search_space = search_space)
   tuner = tnr("irace")
   x = capture.output({tuner$optimize(instance)})
 
@@ -38,27 +42,44 @@ test_that("TunerIrace works with dependencies", {
 })
 
 test_that("TunerIrace works with logical parameters", {
-  search_space = ps(keep_model = p_lgl())
-  instance = TuningInstanceBatchSingleCrit$new(tsk("mtcars"), lrn("regr.rpart"), rsmp("holdout"), msr("regr.mse"),
-    trm("evals", n_evals = 42), search_space)
+  search_space = ps(
+     cp = p_dbl(lower = 0.001, upper = 0.1),
+    keep_model = p_lgl())
+  instance = ti(
+    task = tsk("mtcars"),
+    learner = lrn("regr.rpart"),
+    resampling = rsmp("holdout"),
+    measures = msr("regr.mse"),
+    terminator = trm("evals", n_evals = 96),
+    search_space = search_space)
   tuner = tnr("irace")
   x = capture.output({tuner$optimize(instance)})
   expect_logical(instance$archive$best()$keep_model)
 })
 
 test_that("TunerIrace uses digits", {
   search_space = ps(cp = p_dbl(lower = pi * 1e-20, upper = 5.242e12 / 1e13))
-  instance = TuningInstanceBatchSingleCrit$new(tsk("iris"), lrn("classif.rpart"), rsmp("holdout"), msr("classif.ce"),
-  trm("evals", n_evals = 30), search_space)
+ instance = ti(
+    task = tsk("mtcars"),
+    learner = lrn("regr.rpart"),
+    resampling = rsmp("holdout"),
+    measures = msr("regr.mse"),
+    terminator = trm("evals", n_evals = 96),
+    search_space = search_space)
   tuner = tnr("irace", nbIterations = 1L, minNbSurvival = 1)
   x = capture.output({expect_data_table(tuner$optimize(instance))})
 })
 
 test_that("TunerIrace works with unnamed discrete values", {
   # we had a bug here, see (mlr) issue #627
   search_space = ps(minsplit = p_int(lower = 2L, upper = 7L))
-  inst = TuningInstanceBatchSingleCrit$new(tsk("iris"), lrn("classif.rpart"), rsmp("holdout"), msr("classif.ce"),
-    trm("evals", n_evals = 50), search_space)
+  instance = ti(
+    task = tsk("mtcars"),
+    learner = lrn("regr.rpart"),
+    resampling = rsmp("holdout"),
+    measures = msr("regr.mse"),
+    terminator = trm("evals", n_evals = 96),
+    search_space = search_space)
   tuner = tnr("irace")
-  x = capture.output({expect_data_table(tuner$optimize(inst))})
+  x = capture.output({expect_data_table(tuner$optimize(instance))})
 })