Fix formatting of bag imputation on factors (#800)

* stop with the random ids being printed

* re-cast the predictions due to ipred bug

* fix with of Rnd cli output

* make bagging imputation work with strings_as_factors = FALSE

* undo id specification in Rmd

* better cli options

* update from master for Davis's changes

* ipred version requirement

* udpated docs

* update news

DESCRIPTION

@@ -29,7 +29,7 @@ Imports:
     generics (>= 0.1.0),
     glue,
     gower,
-    ipred,
+    ipred (>= 0.9-12),
     lifecycle,
     lubridate,
     magrittr,

NEWS.md

@@ -25,6 +25,8 @@
 
 * `step_logit()` gained an offset argument for cases where the input is either zero or one (#784)
 
+* A bug was fixed where imputed values via bagged trees would have the wrong levels.
+
 # recipes 0.1.16
 
 ## New Steps

R/impute_bag.R

@@ -180,6 +180,10 @@ step_impute_bag_new <-
 bag_wrap <- function(vars, dat, opt, seed_val) {
   seed_val <- seed_val[1]
   dat <- as.data.frame(dat[, c(vars$y, vars$x)])
+  if (is.character(dat[[vars$y]])) {
+    dat[[vars$y]] <- factor(dat[[vars$y]])
+  }
+
   if (!is.null(seed_val) && !is.na(seed_val))
     set.seed(seed_val)
 
@@ -262,8 +266,8 @@ bake.step_impute_bag <- function(object, new_data, ...) {
         rlang::warn("All predictors are missing; cannot impute")
       } else {
         pred_vals <- predict(object$models[[imp_var]], pred_data)
-        pred_vals <- cast(pred_vals, new_data[[imp_var]])
-        new_data[[imp_var]] <- vec_cast(new_data[[imp_var]], pred_vals)
+        # For an ipred bug reported on 2021-09-14:
+        pred_vals <- cast(pred_vals, object$models[[imp_var]]$y)
         new_data[missing_rows, imp_var] <- pred_vals
       }
     }

man/rmd/recipes.Rmd

@@ -1,5 +1,5 @@
 ```{r startup, include = FALSE}
-options(width = 70)
+options(cli.width = 70, width = 70, cli.unicode = FALSE)
 set.seed(123)
 
 library(dplyr)

tests/testthat/test_impute_bag.R

@@ -100,3 +100,16 @@ test_that('tunable', {
   )
 })
 
+
+test_that('non-factor imputation', {
+  data(scat)
+  scat$Location <- as.character(scat$Location)
+  scat$Location[1] <- NA
+  rec <-
+    recipe(Species ~ ., data = scat) %>%
+    step_impute_bag(Location, impute_with = imp_vars(all_predictors())) %>%
+    prep(strings_as_factors = FALSE)
+  expect_true(is.character(bake(rec, NULL, Location)[[1]]))
+
+})
+