如何高效从数据框列表中按嵌套字符向量列表筛选变量？

Great question—nested loops can get pretty slow when you're dealing with hundreds of variables, so switching to vectorized or functional programming approaches will give you a nice speed boost while keeping your code cleaner. Let's walk through two solid alternatives, one using the purrr package (for readable, concise code) and another using base R (no extra packages needed).

First, let's recap your setup code so we're all on the same page:

# Generate initial data frames
set.seed(1)
dat <- as.data.frame(replicate(n = 8, expr = round(rnorm(3), 2)))
colnames(dat) <- LETTERS[1:8]
dat_list <- list(dat1 = dat, dat2 = dat[, 1:7], dat3 = dat[, 1:4])

# Generate model variable lists
set.seed(1)
colnames_list <- lapply(c(6, 4, 2), function(x) replicate(n = 1, sample(names(dat), size = x, replace = FALSE)))
colnames_list <- lapply(colnames_list, as.vector)
names(colnames_list) <- names(dat_list)
model_list <- list(rpart = colnames_list, lm = colnames_list)

Your original nested loop works, but isn't optimized for large data:

# Original nested loop
subset_list <- list()
for (i in names(model_list)) {
 subset_list[[i]] <- list()
 for (j in names(dat_list)) {
 subset_list[[i]][[j]] <- dat[, model_list[[i]][[j]]]
 }
}

Option 1: Use purrr for clean, fast functional programming #

The purrr package uses vectorized operations under the hood (faster than pure R loops) and makes nested iteration easy to read. We'll use imap to loop through each model (and its name) and map2 to pair each data frame in dat_list with its corresponding variable list:

library(purrr)

subset_list_purrr <- imap(model_list, function(model_cols, model_name) {
  # Pair each data frame with its target columns and subset
  map2(dat_list, model_cols, function(data_df, cols) {
    # drop=FALSE ensures we always get a data frame (not a vector for single columns)
    data_df[, cols, drop = FALSE]
  })
})

Bonus: Handle missing columns (just in case) #

If there's a chance your model variable lists include columns that don't exist in the corresponding data frame (e.g., a column from dat1 that's missing in dat3), add a quick check to only keep existing columns:

subset_list_purrr_safe <- imap(model_list, function(model_cols, model_name) {
  map2(dat_list, model_cols, function(data_df, cols) {
    valid_cols <- intersect(cols, names(data_df))
    data_df[, valid_cols, drop = FALSE]
  })
})

Option 2: Base R with lapply + mapply #

If you don't want to install extra packages, base R's mapply is perfect for pairing dat_list and the variable lists, wrapped in an lapply to iterate over each model:

subset_list_base <- lapply(model_list, function(model_cols) {
  # Pair data frames with their column lists and subset
  mapply(function(data_df, cols) {
    data_df[, cols, drop = FALSE]
  }, dat_list, model_cols, SIMPLIFY = FALSE)
})

SIMPLIFY = FALSE ensures we get a list of data frames instead of a simplified structure, which matches your original output.

Why this is better than nested loops #

Both approaches avoid the overhead of R-level for loops by using vectorized operations that run closer to the C layer. For datasets with hundreds of variables, this will translate to noticeable speed improvements, and the code is more concise and easier to maintain.

内容的提问来源于stack exchange，提问作者markus