如何用高级函数重塑2D矩阵/数据框：合并行列名实现扁平化

Your current implementation with outer() is a clever base R trick! If you're looking for more idiomatic or package-based solutions, here are a few options using popular R tools:

1. Tidyverse (tidyr + dplyr) #

This approach leans into tidy data principles, making each step intuitive and easy to modify:

library(tidyverse)

# Convert matrix to a data frame, keeping row names as a "metric" column
df <- as.data.frame(a) %>%
  mutate(metric = rownames(.))

# Reshape to long format, combine model and metric names, then pivot back to wide
flat_df <- df %>%
  pivot_longer(cols = -metric, names_to = "model") %>%
  unite(col = "model_metric", model, metric, sep = "_") %>%
  pivot_wider(names_from = model_metric, values_from = value)

# Convert to matrix if you need to match your original output structure
flat_matrix <- as.matrix(flat_df)

2. data.table (for fast, memory-efficient reshaping) #

If you’re working with larger datasets, data.table delivers faster reshaping operations with minimal memory overhead:

library(data.table)

# Convert matrix to data.table, preserving row names as a "metric" column
dt <- as.data.table(a, keep.rownames = "metric")

# Melt to long format, then cast back to wide with combined column names
flat_dt <- dcast(
  melt(dt, id.vars = "metric", variable.name = "model"),
  . ~ paste(model, metric, sep = "_"),
  value.var = "value"
)

# Clean up the dummy row identifier and convert to matrix if needed
flat_dt[, . := NULL]
flat_matrix_dt <- as.matrix(flat_dt)

3. Another Base R Method (using stack() and reshape()) #

For a base R alternative that avoids outer(), you can use stacking and built-in reshaping functions:

# Stack the matrix into a long-format data frame
stacked <- stack(a)
# Add the metric (mean/sd) column, repeated for each model
stacked$metric <- rep(rownames(a), ncol(a))
# Combine model and metric into final column names
stacked$model_metric <- paste(stacked$ind, stacked$metric, sep = "_")

# Reshape to wide format
flat_base <- reshape(stacked, idvar = "dummy", timevar = "model_metric", direction = "wide")
# Clean up extra columns and fix names
flat_base <- flat_base[, !names(flat_base) %in% c("dummy", "ind", "metric")]
colnames(flat_base) <- gsub("values\\.", "", colnames(flat_base))
flat_matrix_base <- as.matrix(flat_base)

All these methods will produce the exact flattened structure you need. Your original outer() approach shines for its conciseness, while the tidyverse and data.table options offer better readability and scalability for bigger datasets.

内容的提问来源于stack exchange，提问作者Travis