如何在R中将(n*P)行N列的data.frame转换为n行(N*P)列？

Problem Description #

I'm working with a data.frame df in R that has n*P rows and N columns. The structure looks like this (with row names indicating group-position pairs x-y, where x is the group number (1 to n) and y is the position within the group (1 to P)):

# Example structure (simulated)
set.seed(123)
n <- 3
P <- 4
N <- 4
df <- data.frame(
  C1 = sample(1:200, n*P),
  C2 = sample(-50:100, n*P),
  CN_minus_1 = sample(50:200, n*P),
  CN = sample(10:100, n*P),
  row.names = paste0(rep(1:n, each = P), "-", rep(1:P, n))
)

I need to reshape this into a new data.frame df.new with n rows and N*P columns, following these rules:

• Each row Rk in df.new is formed by column-binding (cbind) all rows from group k in df (i.e., rows k-1, k-2, ..., k-P).
• Each block of N columns in df.new corresponds to the same position across all groups: the first N columns are rows 1-1, 2-1, ..., n-1 row-bound (rbind), the next N columns are rows 1-2, 2-2, ..., n-2 row-bound, and so on until the last N columns.

The target structure should look like this:

# Target structure (conceptual)
# R1: values from df rows 1-1, 1-2, 1-3, 1-P (cbind into one row)
# R2: values from df rows 2-1, 2-2, 2-3, 2-P (cbind into one row)
# Rn: values from df rows n-1, n-2, n-3, n-P (cbind into one row)
# Columns: C1-1, C2-1, ..., CN-1, C1-2, C2-2, ..., CN-2, ..., C1-P, C2-P, ..., CN-P

I tried using nested for loops but it didn't work well, here's my attempt:

for (j in 1:n) {
  df.new <- data.frame(matrix(vector(), 1, dim(df)[2], dimnames = list(c(), colnames(df))), stringsAsFactors=F)
  for (i in 1:nrow(df)) {
    if (i %% n == 0) {
      df.new <- rbind(df.new, df[i,])
    } else if (i %% n == j) {
      df.new <- rbind(df.new, df[i,])
    }
  }
  assign(paste0("df.new", j), df.new)
}

Solution #

Your loop approach is inefficient because repeated rbind operations copy the entire data frame each time, which gets slow with large datasets. Instead, use vectorized operations or tidyverse tools for a cleaner, faster solution.

Method 1: Base R (No External Packages)

This approach leverages matrix transposition and list operations to reshape the data efficiently:

# Step 1: Split the original data into n groups (each with P rows)
group <- rep(1:n, each = P)
df_groups <- split(df, group)

# Step 2: Transpose each group to turn P rows into 1 row, then combine all groups
df.new <- do.call(rbind, lapply(df_groups, function(g) {
  as.data.frame(t(g), stringsAsFactors = FALSE)
}))

# Step 3: Set meaningful column and row names
colnames(df.new) <- paste0(rep(colnames(df), P), "-", rep(1:P, each = N))
rownames(df.new) <- paste0("R", 1:n)

Method 2: Tidyverse (dplyr + tidyr)

If you prefer a more readable, pipe-based workflow, use the tidyverse packages:

library(dplyr)
library(tidyr)

# Step 1: Add group and position columns to the original data
df_with_groups <- df %>%
  mutate(
    group = rep(1:n, each = P),
    pos = rep(1:P, n)
  )

# Step 2: Reshape from wide to long, then back to wide in the desired format
df.new <- df_with_groups %>%
  pivot_longer(cols = starts_with("C"), names_to = "column", values_to = "value") %>%
  pivot_wider(
    id_cols = group,
    names_from = c(column, pos),
    values_from = value,
    names_sep = "-"
  ) %>%
  column_to_rownames("group")

# Step 3: Set row names (matches your target structure)
rownames(df.new) <- paste0("R", 1:n)

Both methods will produce the exact structure you need, with better performance and maintainability than nested loops.

内容的提问来源于stack exchange，提问作者Makoto Miyazaki