type checks fail uninformatively when outcome has "label" attribute

[mesdi]

library(tidyverse)
library(WDI)

#Dataset for methane emissions
df_methane <- 
  WDI(indicator = "EN.ATM.METH.KT.CE", 
      extra = TRUE) %>% 
  as_tibble() %>% 
  janitor::clean_names() %>% 
  drop_na() %>% 
  rename(methane = en_atm_meth_kt_ce) 

#Modeling
library(tidymodels)

#Dataset for modeling
df_mod <- 
  df_methane %>% 
  filter(region != "Aggregates",
         income != "Not classified") %>% 
  mutate(latitude = latitude %>% as.numeric(),
         longitude = longitude %>% as.numeric(),
         income = income %>% as_factor(),
         region = region %>% as_factor()) %>% 
  select(region, income, longitude,latitude, methane) 


#Splitting
set.seed(12345)
df_split <- initial_split(df_mod, 
                          prop = 0.8,
                          strata = "income")

df_train <- training(df_split)
df_test <- testing(df_split)

tidymodels/workflows#10-fold cross validation for tuning
set.seed(12345)
df_fold <- 
  vfold_cv(df_train,
           strata = income,
           repeats = 5)

#Linear regression models for different engines/packages
spec_lm <- 
  linear_reg() %>% 
  set_engine("lm") 


spec_glm <- 
  linear_reg() %>% 
  set_engine("glm")

spec_glmnet <- 
  linear_reg(penalty = tune(),
             mixture = tune()) %>% 
  set_engine("glmnet")

spec_keras <- 
  linear_reg(penalty = tune()) %>% 
  set_engine("keras")

spec_stan <- 
  linear_reg() %>% 
  set_engine("stan")

#Workflow set
basic_recipe <-
  recipe(methane ~ ., data = df_train) 

all_workflows <- 
  workflow_set(
    preproc = list(basic = basic_recipe), 
    models = list(LM = spec_lm, 
                  GLM = spec_glm,
                  GLMNET = spec_glmnet,
                  Keras = spec_keras,
                  Stan = spec_stan)
  )

#Tuning and evaluating the models
grid_ctrl <-
  control_grid(
    save_pred = TRUE,
    parallel_over = "everything",
    save_workflow = TRUE
  )

grid_results <-
  all_workflows %>%
  workflow_map(
    seed = 98765,
    resamples = df_fold,
    grid = 15,
    control = grid_ctrl
  )

#one of the error lines:  
# `y` should be one of the following classes: 'data.frame', 'matrix', 'factor', 'Surv'

[simonpcouch]

Ah, this was an interesting one. Thanks for the issue!

I'm able to reproduce just passing a linear regression specification to tune_grid(). The issue here is that, when fitting regression models, tidymodels usually expects the outcome to be a vector. It isn't, though:

library(tidyverse)
library(tidymodels)
library(WDI)

df_methane <- 
   WDI(indicator = "EN.ATM.METH.KT.CE", 
       extra = TRUE) %>% 
   as_tibble() %>% 
   janitor::clean_names() %>% 
   drop_na() %>% 
   rename(methane = en_atm_meth_kt_ce) 

is.vector(df_methane$methane)
#> [1] FALSE

It's not a vector because WDI() has attached a "label" attribute to it:

attr(df_methane$methane, "label")
#> [1] "Methane emissions (kt of CO2 equivalent)"

Thus:

df_mod <- 
   df_methane %>% 
   filter(region != "Aggregates",
          income != "Not classified") %>% 
   mutate(latitude = latitude %>% as.numeric(),
          longitude = longitude %>% as.numeric(),
          income = income %>% as_factor(),
          region = region %>% as_factor()) %>% 
   select(region, income, longitude,latitude, methane) 

set.seed(12345)
df_split <- initial_split(df_mod, 
                          prop = 0.8,
                          strata = "income")

df_train <- training(df_split)
df_test <- testing(df_split)

set.seed(12345)
df_fold <- 
   vfold_cv(df_train,
            strata = income,
            repeats = 5)

res <-
   tune_grid(
      linear_reg(),
      methane ~ ., 
      df_fold
   )
#> Warning: No tuning parameters have been detected, performance will be evaluated
#> using the resamples with no tuning. Did you want to [tune()] parameters?
#> → A | error:   `y` should be one of the following classes: 'data.frame', 'matrix', 'factor', 'Surv'
#> There were issues with some computations   A: x1

Your solution is to remove the label attribute:

# remove the label attribute ---------------------------------------------
attr(df_methane$methane, "label") <- NULL

df_mod <- 
   df_methane %>% 
   filter(region != "Aggregates",
          income != "Not classified") %>% 
   mutate(latitude = latitude %>% as.numeric(),
          longitude = longitude %>% as.numeric(),
          income = income %>% as_factor(),
          region = region %>% as_factor()) %>% 
   select(region, income, longitude,latitude, methane) 

set.seed(12345)
df_split <- initial_split(df_mod, 
                          prop = 0.8,
                          strata = "income")

df_train <- training(df_split)
df_test <- testing(df_split)

set.seed(12345)
df_fold <- 
   vfold_cv(df_train,
            strata = income,
            repeats = 5)


res <-
   tune_grid(
      linear_reg(),
      methane ~ ., 
      df_fold
   )
#> Warning: No tuning parameters have been detected, performance will be evaluated
#> using the resamples with no tuning. Did you want to [tune()] parameters?

^{Created on 2024-02-02 with reprex v2.1.0}

I'm going to move this to parsnip, where the issue originates.

[simonpcouch]

The offending lines:

parsnip/R/fit.R

Lines 414 to 417 in 8ccf1be

	# `y` can be a vector (which is not a class), or a factor or
	# Surv object (which are not vectors)
	if (!is.null(y) && !is.vector(y))
	inher(y, c("data.frame", "matrix", "factor", "Surv"), cl)

[simonpcouch]

Here's a more minimal reprex:

library(parsnip)

fit_xy(
  linear_reg(),
  data.frame(x = 1:5),
  y = data.frame(structure(rnorm(5), label = "hi"))
)
#> Error in `inher()`:
#> ! `y` should be one of the following classes: 'data.frame', 'matrix', 'factor', 'Surv'

^{Created on 2024-02-02 with reprex v2.1.0}

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.