check tidyselect output in fit.tailor()

R/tailor.R

@@ -86,13 +86,13 @@ fit.tailor <- function(object, .data, outcome, estimate, probabilities = c(),
 
   columns <- list()
   columns$outcome <- names(tidyselect::eval_select(enquo(outcome), .data))
+  check_selection(enquo(outcome), columns$outcome, "outcome")
   columns$estimate <- names(tidyselect::eval_select(enquo(estimate), .data))
-
-  probabilities <- tidyselect::eval_select(enquo(probabilities), .data)
-  if (length(probabilities) > 0) {
-    columns$probabilities <- names(probabilities)
-  } else {
-    columns$probabilities <- character(0)
+  check_selection(enquo(estimate), columns$estimate, "estimate")
+  columns$probabilities <- names(tidyselect::eval_select(enquo(probabilities), .data))
+  if (any(c("probability", "everything") %in%
+          purrr::map_chr(object$adjustments, purrr::pluck, "inputs"))) {
+    check_selection(enquo(probabilities), columns$probabilities, "probabilities")
   }
 
   time <- tidyselect::eval_select(enquo(time), .data)

R/utils.R

@@ -176,3 +176,15 @@ check_method <- function(method,
   method
 }
 
+check_selection <- function(selector, result, arg, call = caller_env()) {
+  if (length(result) == 0) {
+    cli_abort(
+      c(
+        "!" = "{.arg {arg}} must select at least one column.",
+        "x" = "Selector {.code {as_label(selector)}} did not match any columns \\
+               in {.arg .data}."
+      ),
+      call = caller_env()
+    )
+  }
+}

tests/testthat/_snaps/tailor.md

@@ -40,3 +40,58 @@
       * Adjust probability threshold to 0.2.
       * Add equivocal zone of size 0.1.
 
+# error informatively with empty tidyselections
+
+    Code
+      tailor_fit <- tailor() %>% adjust_probability_threshold(0.5) %>% fit(
+        two_class_example, outcome = "truth_WRONG", estimate = "predicted",
+        probabilities = tidyselect::contains("Class"))
+    Condition
+      Error in `fit()`:
+      ! Can't select columns that don't exist.
+      x Column `truth_WRONG` doesn't exist.
+
+---
+
+    Code
+      tailor_fit <- tailor() %>% adjust_probability_threshold(0.5) %>% fit(
+        two_class_example, outcome = contains("truth_WRONG"), estimate = "predicted",
+        probabilities = tidyselect::contains("Class"))
+    Condition
+      Error in `fit()`:
+      ! `outcome` must select at least one column.
+      x Selector `contains("truth_WRONG")` did not match any columns in `.data`.
+
+---
+
+    Code
+      tailor_fit <- tailor() %>% adjust_probability_threshold(0.5) %>% fit(
+        two_class_example, outcome = "truth", estimate = "predicted_WRONG",
+        probabilities = tidyselect::contains("Class"))
+    Condition
+      Error in `fit()`:
+      ! Can't select columns that don't exist.
+      x Column `predicted_WRONG` doesn't exist.
+
+---
+
+    Code
+      tailor_fit <- tailor() %>% adjust_probability_threshold(0.5) %>% fit(
+        two_class_example, outcome = "truth", estimate = contains("predicted_WRONG"),
+        probabilities = tidyselect::contains("Class"))
+    Condition
+      Error in `fit()`:
+      ! `estimate` must select at least one column.
+      x Selector `contains("predicted_WRONG")` did not match any columns in `.data`.
+
+---
+
+    Code
+      tailor_fit <- tailor() %>% adjust_probability_threshold(0.5) %>% fit(
+        two_class_example, outcome = contains("truth"), estimate = "predicted",
+        probabilities = tidyselect::contains("Class_WRONG"))
+    Condition
+      Error in `fit()`:
+      ! `probabilities` must select at least one column.
+      x Selector `tidyselect::contains("Class_WRONG")` did not match any columns in `.data`.
+

tests/testthat/test-tailor.R

@@ -11,3 +11,100 @@ test_that("tailor printing", {
       adjust_equivocal_zone()
   )
 })
+
+test_that("error informatively with empty tidyselections", {
+  skip_if_not_installed("modeldata")
+  data("two_class_example", package = "modeldata")
+
+  expect_no_condition(
+    tailor_fit <- tailor() %>%
+      adjust_probability_threshold(.5) %>%
+      fit(
+        two_class_example,
+        outcome = "truth",
+        estimate = "predicted",
+        probabilities = tidyselect::contains("Class")
+      )
+  )
+
+  # outcome doesn't exist, is bare string
+  expect_snapshot(
+    error = TRUE,
+    tailor_fit <- tailor() %>%
+      adjust_probability_threshold(.5) %>%
+      fit(
+        two_class_example,
+        outcome = "truth_WRONG",
+        estimate = "predicted",
+        probabilities = tidyselect::contains("Class")
+      )
+  )
+
+  # outcome doesn't exist, is selection helper
+  expect_snapshot(
+    error = TRUE,
+    tailor_fit <- tailor() %>%
+      adjust_probability_threshold(.5) %>%
+      fit(
+        two_class_example,
+        outcome = contains("truth_WRONG"),
+        estimate = "predicted",
+        probabilities = tidyselect::contains("Class")
+      )
+  )
+
+  # estimate doesn't exist, is bare string
+  expect_snapshot(
+    error = TRUE,
+    tailor_fit <- tailor() %>%
+      adjust_probability_threshold(.5) %>%
+      fit(
+        two_class_example,
+        outcome = "truth",
+        estimate = "predicted_WRONG",
+        probabilities = tidyselect::contains("Class")
+      )
+  )
+
+  # estimate doesn't exist, is selection helper
+  expect_snapshot(
+    error = TRUE,
+    tailor_fit <- tailor() %>%
+      adjust_probability_threshold(.5) %>%
+      fit(
+        two_class_example,
+        outcome = "truth",
+        estimate = contains("predicted_WRONG"),
+        probabilities = tidyselect::contains("Class")
+      )
+  )
+
+  # probability doesn't exist, is selection helper, is needed
+  expect_snapshot(
+    error = TRUE,
+    tailor_fit <- tailor() %>%
+      adjust_probability_threshold(.5) %>%
+      fit(
+        two_class_example,
+        outcome = contains("truth"),
+        estimate = "predicted",
+        probabilities = tidyselect::contains("Class_WRONG")
+      )
+  )
+
+  # probability doesn't exist, is selection helper, isn't needed
+  # (asserting here that we ought not to error on a bad selection
+  # if it would not be used anyway.)
+  # todo: need to overwrite column name for now, see #22.
+  two_class_example$.pred <- two_class_example$Class2
+  expect_no_condition(
+    tailor_fit <- tailor() %>%
+      adjust_numeric_range(.5) %>%
+      fit(
+        two_class_example,
+        outcome = "Class1",
+        estimate = ".pred",
+        probabilities = tidyselect::contains("Class_WRONG")
+      )
+  )
+})

tests/testthat/test-utils.R

@@ -17,7 +17,8 @@ test_that("tailor_fully_trained works", {
         fit(
           two_class_example,
           outcome = "truth",
-          estimate = tidyselect::contains("Class")
+          estimate = tidyselect::contains("Class"),
+          probabilities = tidyselect::contains("Class")
         ) %>%
         adjust_probability_threshold(.5)
     )
@@ -33,7 +34,8 @@ test_that("tailor_fully_trained works", {
           # todo: this function requires a different format of `estimate`
           # and `probabilities` specification than the call below to
           # be able to fit properly.
-          estimate = tidyselect::contains("Class")
+          estimate = tidyselect::contains("Class"),
+          probabilities = tidyselect::contains("Class")
         )
     )
   )