【问题标题】：dplyr: summarize long format for multiple groupsdplyr：汇总多个组的长格式
【发布时间】：2019-07-10 09:56:05
【问题描述】：

我知道有很多问题，听起来可能有些相似，但我无法找到我确切问题的答案。

假设我们有一个玩具数据集

library(tidyverse)
df <- tibble(
  Gender = c("m", "f", "f", "m", "m", 
             "f", "f", "f", "m", "f"),
  IQ = rnorm(10, 100, 15),
  Other = runif(10),
  Test = rnorm(10),
  group2 = c("A", "A", "A", "A", "A",
             "B", "B", "B", "B", "B")
)

我们要从中计算mean、min 和max 的gender 和group2。

只有一组，我可以很容易地写出来

df %>% 
  group_by(Gender) %>% 
  select_if(is.numeric) %>% 
  gather(Variable, Value, -Gender) %>% 
  group_by(Variable, Gender) %>% 
  summarise(mean = mean(Value), 
        min = min(Value), 
        max = max(Value)) %>% 
 ungroup()

得到

 Variable Gender    mean     min     max
 <chr>    <chr>    <dbl>   <dbl>   <dbl>
1 IQ       f      99.2    81.9    121.   
2 IQ       m      89.0    62.5    106.   
3 Other    f       0.301   0.187    0.479
4 Other    m       0.395   0.0483   0.757
5 Test     f      -0.0770 -1.18     0.545
6 Test     m       0.163  -0.632    0.828

但我不知道如何为多个组做同样的事情。我知道我可以像这样使用summarise_*()

df %>% 
  group_by(Gender) %>% 
  summarise_if(is.numeric, list(mean = mean, 
                                min = min, 
                                max = max))

但它返回宽格式（如data.table）

  Gender IQ_mean Other_mean Test_mean IQ_min Other_min Test_min IQ_max
  <chr>    <dbl>      <dbl>     <dbl>  <dbl>     <dbl>   <dbl>  <dbl>
1 f         99.2      0.301   -0.0770   81.9    0.187   -1.18    121.
2 m         89.0      0.395    0.163    62.5    0.0483  -0.632   106.
# … with 2 more variables: Other_max <dbl>, Test_max <dbl>

当您有 10 多个变量时，这几乎没有用。

我在这里缺少什么？

【问题讨论】：

标签： r dplyr data.table

【解决方案1】：

您可以在自己的代码中添加gather、separate 和spread：

df %>% 
    group_by(Gender, group2) %>% 
    summarise_if(is.numeric, list(mean = mean, 
                                  min = min, 
                                  max = max)) %>% 
    gather(vars, vals, -Gender, -group2) %>% 
    separate(vars, c("Variable", "stat")) %>% 
    spread(stat, vals)

#### OUTPUT ####

# A tibble: 12 x 6
# Groups:   Gender [2]
   Gender group2 Variable     max    mean       min
   <chr>  <chr>  <chr>      <dbl>   <dbl>     <dbl>
 1 f      A      IQ       110.    103.     95.0    
 2 f      A      Other      0.934   0.469   0.00439
 3 f      A      Test       1.39    0.472  -0.446  
 4 f      B      IQ       121.     92.0    75.6    
 5 f      B      Other      0.730   0.461   0.261  
 6 f      B      Test       0.589   0.276  -0.524  
 7 m      A      IQ       112.    104.     94.3    
 8 m      A      Other      0.827   0.613   0.308  
 9 m      A      Test       0.724   0.136  -0.264  
10 m      B      IQ       115.    115.    115.     
11 m      B      Other      0.970   0.970   0.970  
12 m      B      Test      -1.05   -1.05   -1.05

【讨论】：

【解决方案2】：

您可以先将IQ、Other 和Test 收集在单个变量列中，将df 转换为长格式，然后计算每个组的汇总统计信息（Gender-group2-variable）：

library(tidyverse)

set.seed(1)

## data
df <- tibble(
    Gender = c("m", "f", "f", "m", "m", 
        "f", "f", "f", "m", "f"),
    IQ = rnorm(10, 100, 15),
    Other = runif(10),
    Test = rnorm(10),
    group2 = c("A", "A", "A", "A", "A",
        "B", "B", "B", "B", "B")
)

df %>%
    gather(key = "variable", value = "value", -c(Gender, group2)) %>%
    group_by(Gender, group2, variable) %>%
    summarize_at("value", list(mean = mean, min = min, max = max)) %>%
    ungroup()
#> # A tibble: 12 x 6
#>    Gender group2 variable    mean      min     max
#>    <chr>  <chr>  <chr>      <dbl>    <dbl>   <dbl>
#>  1 f      A      IQ        95.1    87.5    103.   
#>  2 f      A      Other      0.432   0.212    0.652
#>  3 f      A      Test       0.464  -0.0162   0.944
#>  4 f      B      IQ       100.     87.7    111.   
#>  5 f      B      Other      0.281   0.0134   0.386
#>  6 f      B      Test       0.599   0.0746   0.919
#>  7 m      A      IQ       106.     90.6    124.   
#>  8 m      A      Other      0.442   0.126    0.935
#>  9 m      A      Test       0.457  -0.0449   0.821
#> 10 m      B      IQ       109.    109.     109.   
#> 11 m      B      Other      0.870   0.870    0.870
#> 12 m      B      Test      -1.99   -1.99    -1.99

【讨论】：

你知道用这种方法在自己的函数中添加任意数量的组吗？ ... 似乎没有奏效。编辑：最好只是提出一个新问题，谢谢！
我不太确定您指的是哪个函数的点，您能否进一步澄清一下？
见这里stackoverflow.com/questions/56968968/…

【解决方案3】：

这是data.table 方法

library( data.table )
melt( setDT(df), 
  id.vars = c("Gender", "group2") )[, .(max = max(value, na.rm = TRUE), 
                                        min = min(value, na.rm = TRUE),
                                        mean = mean(value, na.rm = TRUE)),
                                    by = .(Gender, group2, variable )][]

#    Gender group2 variable           max          min         mean
# 1:      m      A       IQ 120.739562935  83.46037366  96.99412720
# 2:      f      A       IQ  98.657598754  98.43677811  98.54718843
# 3:      f      B       IQ 111.973534436  71.38605822  94.04719457
# 4:      m      B       IQ 102.913093964 102.91309396 102.91309396
# 5:      m      A    Other   0.861929066   0.51651983   0.66098944
# 6:      f      A    Other   0.752484881   0.07648229   0.41448359
# 7:      f      B    Other   0.463524836   0.18308752   0.33301693
# 8:      m      B    Other   0.099740011   0.09974001   0.09974001
# 9:      m      A     Test   1.159379020  -0.83569116   0.04268551
# 10:      f      A     Test  -0.009017293  -0.77245300  -0.39073515
# 11:      f      B     Test   1.591132150  -0.99248570  -0.24997246
# 12:      m      B     Test   1.654489766   1.65448977   1.65448977

基准

# Unit: milliseconds
#       expr       min        lq      mean    median        uq       max neval
# data.table  1.498788  1.819936  1.997320  1.980358  2.218809  2.413124    10
# tidyverse1 11.263956 11.887270 12.421442 11.963340 12.484075 15.401816    10
# tidyverse2  4.952477  5.185053  6.303103  6.001478  6.902558  9.663341    10

microbenchmark::microbenchmark(
  data.table = {
    DT <- copy(df)
    melt( setDT(DT), 
          id.vars = c("Gender", "group2") )[, .(max = max(value, na.rm = TRUE), 
                                                min = min(value, na.rm = TRUE),
                                                mean = mean(value, na.rm = TRUE)),
                                            by = .(Gender, group2, variable )][]

  },
  tidyverse1 = {
    DT <- copy(df)
    df %>% 
      group_by(Gender, group2) %>% 
      summarise_if(is.numeric, list(mean = mean, 
                                    min = min, 
                                    max = max)) %>% 
      gather(vars, vals, -Gender, -group2) %>% 
      separate(vars, c("Variable", "stat")) %>% 
      spread(stat, vals)
  },
  tidyverse2 = {
    df %>%
      gather(key = "variable", value = "value", -c(Gender, group2)) %>%
      group_by(Gender, group2, variable) %>%
      summarize_at("value", list(mean = mean, min = min, max = max)) %>%
      ungroup()
  },
  times = 10 
)

【讨论】：