dplyr中Mutate调用触发深层"Evaluation Error"问题求助

Hey there! Let's break down why your 1-of-(C-1) effect coding function works standalone but throws an Evaluation Error when used with mutate(), plus actionable fixes to get it running smoothly with your Pokémon dataset:

Common Culprits & Fixes #

1. Your function isn't vectorized for dplyr's workflow #

mutate() passes entire columns (vectors) to functions, but if your function only handles single values, it'll break. For example, if you wrote it to take one category value and return a vector, feeding it a full column of values will cause mismatched lengths or logic errors.

Fix: Use purrr::map or rowwise() to apply your function to each row individually, and wrap the output in a list to store vector results in a column:

library(dplyr)
library(purrr)

# Example vector-friendly effect coding function
effect_encode <- function(x, ref_level) {
  all_levels <- unique(na.omit(x))
  coding_levels <- setdiff(all_levels, ref_level)
  n_codes <- length(coding_levels)
  
  # Use map to handle each value in the input vector
  map(x, function(val) {
    if (is.na(val)) {
      rep(NA, n_codes)
    } else if (val == ref_level) {
      rep(-1, n_codes)
    } else {
      vec <- rep(-1, n_codes)
      vec[which(coding_levels == val)] <- 1
      vec
    }
  })
}

# Use in mutate()
pokemon_df %>%
  mutate(type_1_effect = effect_encode(type_1, ref_level = "Water"))

2. Non-standard evaluation (NSE) conflicts with dplyr #

dplyr uses NSE to reference columns, which can cause issues if your function relies on global variables or doesn't explicitly reference columns with dplyr's .data pronoun.

Fix: Pass all required parameters explicitly to your function, and use .data$col_name if you need to reference columns inside the function to avoid environment confusion.

3. Unhandled edge cases in your dataset #

Your standalone test might use clean, expected values, but your Pokémon dataset could have NAs, unexpected category levels, or factor/character mismatches that break the function.

Fix:

• Add NA handling logic (like in the example function above)
• Verify your category columns are the right type (check with glimpse(pokemon_df))
• Ensure your function accounts for all levels present in the column (not just the ones you tested)

4. Use your traceback() output to pinpoint the error #

Since you already ran traceback(), look at the lowest-level error message:

• If it mentions "length mismatch", your function is returning vectors of varying lengths for different rows
• If it says "object not found", you're referencing a variable that dplyr can't locate in its evaluation environment
• If it's a logical error (e.g., which() returning an empty vector), your function isn't handling a category level present in the dataset

Full Working Example #

Here's a complete snippet tailored to Pokémon data:

# Simulate a sample Pokémon dataset
pokemon_df <- tibble(
  type_1 = sample(c("Fire", "Water", "Grass", "Electric"), 100, replace = TRUE),
  type_2 = sample(c(NA, "Flying", "Poison"), 100, replace = TRUE),
  attack = rnorm(100, 80, 20)
)

# Apply effect coding
pokemon_processed <- pokemon_df %>%
  mutate(
    type_1_effect = effect_encode(type_1, ref_level = "Water"),
    type_2_effect = effect_encode(type_2, ref_level = "Flying")
  )

# Inspect results
pokemon_processed %>% select(type_1, type_1_effect) %>% head()

内容的提问来源于stack exchange，提问作者Matthew Wolff