finalize parameters with unknown ranges in tune_sim_anneal()

tidymodels · Jan 19, 2024 · dfcd6d7 · dfcd6d7
1 parent 4d0fda5
commit dfcd6d7
Show file tree

Hide file tree

Showing 4 changed files with 48 additions and 18 deletions.
diff --git a/NEWS.md b/NEWS.md
@@ -1,5 +1,7 @@
 # finetune (development version)
 
+* Fixed bug where `tune_sim_anneal()` would fail when supplied parameters needing finalization. The function will now finalize needed parameter ranges internally (#39).
+
 * `autoplot()` methods for racing objects will now use integers in x-axis breaks (#75).
 
 * Enabling the `verbose_elim` control option for `tune_race_anova()` will now additionally introduce a message confirming that the function is evaluating against the burn-in resamples.

diff --git a/R/tune_sim_anneal.R b/R/tune_sim_anneal.R
@@ -328,12 +328,12 @@ tune_sim_anneal_workflow <-
     }
 
     # In case mtry or other parameters with missing ranges need finalization
-    # param_info <- tune::check_parameters(
-    #   workflow = object,
-    #   pset = param_info,
-    #   data = resamples$splits[[1]]$data,
-    #   grid_names = param_info$id
-    # )
+    param_info <- tune::check_parameters(
+      wflow = object,
+      pset = param_info,
+      data = resamples$splits[[1]]$data,
+      grid_names = character(0)
+    )
 
     tune::check_workflow(object, check_dials = !is.null(param_info), pset = param_info)
 

diff --git a/tests/testthat/_snaps/sa-overall.md b/tests/testthat/_snaps/sa-overall.md
@@ -205,16 +205,37 @@
       set.seed(40)
       rf_res_finetune <- wf_rf %>% tune_sim_anneal(resamples = bt, initial = rf_res)
     Message
+      i Creating pre-processing data to finalize unknown parameter: mtry
       Optimizing roc_auc
-      Initial best: 0.86248
-      1 ( ) accept suboptimal  roc_auc=0.86132 (+/-0.007045)
-      2 ( ) accept suboptimal  roc_auc=0.85987 (+/-0.007598)
-      3 + better suboptimal  roc_auc=0.86155 (+/-0.007228)
-      4 + better suboptimal  roc_auc=0.86193 (+/-0.007212)
-      5 ( ) accept suboptimal  roc_auc=0.86108 (+/-0.00727)
-      6 ( ) accept suboptimal  roc_auc=0.85992 (+/-0.007641)
-      7 + better suboptimal  roc_auc=0.86045 (+/-0.007405)
-      8 x restart from best  roc_auc=0.85863 (+/-0.007596)
-      9 ( ) accept suboptimal  roc_auc=0.8623 (+/-0.007265)
-      10 <3 new best           roc_auc=0.86273 (+/-0.007569)
+      Initial best: 0.84856
+      1 ( ) accept suboptimal  roc_auc=0.84375 (+/-0.007727)
+      2 <3 new best           roc_auc=0.84943 (+/-0.007036)
+      3 ( ) accept suboptimal  roc_auc=0.84371 (+/-0.007903)
+      4 + better suboptimal  roc_auc=0.84825 (+/-0.008036)
+      5 ( ) accept suboptimal  roc_auc=0.84479 (+/-0.00814)
+      6 + better suboptimal  roc_auc=0.84816 (+/-0.007283)
+      7 ( ) accept suboptimal  roc_auc=0.84381 (+/-0.007999)
+      8 <3 new best           roc_auc=0.85014 (+/-0.007172)
+      9 ( ) accept suboptimal  roc_auc=0.84344 (+/-0.007818)
+      10 + better suboptimal  roc_auc=0.84802 (+/-0.007281)
+
+---
+
+    Code
+      set.seed(40)
+      rf_res_finetune <- wf_rf %>% tune_sim_anneal(resamples = bt)
+    Message
+      i Creating pre-processing data to finalize unknown parameter: mtry
+      Optimizing roc_auc
+      Initial best: 0.84829
+      1 ( ) accept suboptimal  roc_auc=0.84326 (+/-0.007765)
+      2 <3 new best           roc_auc=0.84843 (+/-0.007903)
+      3 ( ) accept suboptimal  roc_auc=0.84438 (+/-0.007904)
+      4 <3 new best           roc_auc=0.84881 (+/-0.007951)
+      5 ( ) accept suboptimal  roc_auc=0.84404 (+/-0.007787)
+      6 <3 new best           roc_auc=0.84912 (+/-0.007272)
+      7 ( ) accept suboptimal  roc_auc=0.84333 (+/-0.007685)
+      8 + better suboptimal  roc_auc=0.84867 (+/-0.007152)
+      9 - discard suboptimal roc_auc=0.84525 (+/-0.007793)
+      10 ( ) accept suboptimal  roc_auc=0.84383 (+/-0.00773)
 
diff --git a/tests/testthat/test-sa-overall.R b/tests/testthat/test-sa-overall.R
@@ -101,7 +101,7 @@ test_that("unfinalized parameters", {
   rec_example <- recipe(Class ~ ., data = two_class_dat)
 
   # RF
-  model_rf <- rand_forest(min_n = tune()) %>%
+  model_rf <- rand_forest(mtry = tune()) %>%
     set_mode("classification") %>%
     set_engine("ranger")
 
@@ -118,6 +118,13 @@ test_that("unfinalized parameters", {
     rf_res_finetune <- wf_rf %>%
       tune_sim_anneal(resamples = bt, initial = rf_res)
   })
+
+  # don't supply an initial grid (#39)
+  expect_snapshot({
+    set.seed(40)
+    rf_res_finetune <- wf_rf %>%
+      tune_sim_anneal(resamples = bt)
+  })
 })
 
 test_that("set event-level", {