在R中实现两级拆分：将数据框转为双层列表向量

问题背景 #

现有如下结构的data.frame h：

component     hyperpar       x
1          i       sigma2     1.0
2       envN       sigma2     1.0
3       envN   sigma2_int     0.1
4       envN   sigma2_int     0.1
...
85        dt       sigma2     1.0
86        dt           ls 40000.0
87        dt           ls 40000.0
88      year       sigma2     1.0

需要将其转换为双层列表：外层列表以component值为名称，内层列表以hyperpar值为名称，对应的值是该分组下的x向量。

已通过unstack(h, x ~ component)实现单层拆分，但尝试unstack(h, x ~ component ~ hyperpar)或unstack(h, x ~ component + hyperpar)均未得到预期的双层结构，希望用tidyverse的简洁方案实现。

完整测试数据集：

h <- structure(list(component = c("i", "envN", "envN", "envN", "envN", 
"envN", "envN", "envN", "envN", "envN", "envN", "envN", "envN", 
"envN", "envN", "envN", "envN", "envN", "envN", "envN", "envN", 
"envN", "envN", "envN", "envN", "envN", "envN", "envN", "envN", 
"envN", "envN", "envN", "envN", "envN", "envN", "envN", "envN", 
"envN", "envN", "envN", "envN", "envN", "envN", "envN", "envN", 
"envN", "envN", "envN", "envN", "envN", "envN", "envN", "envN", 
"envN", "env", "env", "env", "env", "env", "env", "env", "env", 
"env", "env", "env", "env", "env", "env", "env", "env", "env", 
"env", "env", "env", "env", "env", "env", "env", "env", "env", 
"spat", "spat", "spat", "dt", "dt", "dt", "year"), hyperpar = c("sigma2", 
"sigma2", "sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", 
"sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", 
"sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", 
"sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", 
"sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", "sigma2_int", 
"sigma2_int", "sigma2_int", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2_slope", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2_slope", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2_slope", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2_slope", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2_slope", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2_slope", "sigma2_slope", "sigma2_slope", "sigma2_slope", 
"sigma2", "ls", "ls", "ls", "ls", "ls", "ls", "ls", "ls", "ls", 
"ls", "ls", "ls", "ls", "ls", "ls", "ls", "ls", "ls", "ls", "ls", 
"ls", "ls", "ls", "ls", "ls", "ls", "ls", "sigma2", "ls", "ls", 
"sigma2", "ls", "ls", "sigma2"), x = c(1, 1, 0.1, 0.1, 0.1, 0.1, 
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 10, 10, 10, 
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 
10, 10, 10, 10, 10, 10, 10, 10, 1, 40000, 40000, 40000, 40000, 
40000, 40000, 40000, 40000, 40000, 40000, 40000, 40000, 40000, 
40000, 40000, 40000, 40000, 40000, 40000, 40000, 40000, 40000, 
40000, 40000, 40000, 40000, 1, 10, 10, 1, 40000, 40000, 1)), row.names = c(NA, 
-88L), class = "data.frame")

解决方案 #

方法一：分组嵌套 + 映射转换 #

利用dplyr的分组、嵌套功能，结合purrr的映射函数，逐步构建双层列表：

library(tidyverse)

result <- h %>%
  # 先按component和hyperpar分组，将每组的x转为列表
  group_by(component, hyperpar) %>%
  summarise(x_vec = list(x), .groups = "drop") %>%
  # 再按component分组，嵌套hyperpar和x_vec
  group_by(component) %>%
  nest() %>%
  # 将每个component内的嵌套数据框转为hyperpar为名称的列表
  mutate(data = map(data, ~deframe(.x))) %>%
  # 最终转为以component为名称的双层列表
  deframe()

查看结果示例：

# 查看envN对应的内层列表
result$envN
# $sigma2
# [1] 1
# 
# $sigma2_int
#  [1] 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
# 
# $sigma2_slope
#  [1] 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10

方法二：分组内直接unstack #

更简洁的写法，在每个component分组内用unstack实现内层拆分：

result2 <- h %>%
  group_by(component) %>%
  # 对每个component分组内的数据，用unstack拆分出hyperpar层级的列表
  summarise(hyperpar_list = list(unstack(cur_data(), x ~ hyperpar))) %>%
  # 转为双层列表
  deframe()

这个方法逻辑更直接，利用cur_data()获取当前分组内的数据，再通过unstack生成内层列表，最后用deframe整理为外层以component为名称的双层结构。

两种方法都能得到符合需求的双层列表，且完全遵循tidyverse的链式语法风格，避免了繁琐的lapply嵌套写法。

内容的提问来源于stack exchange，提问作者Tomas