跨不同數據幀的條件匹配列

Question

我正在使用兩個數據集 - 一組有成對的項目：

original <- data.frame(label1 = c("cat", "cat", "dog", "dog", "cat", "tiger", "tiger", "cow"),
                      label2 = c("dog", "dog", "cat", "cat", "dog", "cow", "cow", "tiger"))
original
  label1 label2
1    cat    dog
2    cat    dog
3    dog    cat
4    dog    cat
5    cat    dog
6  tiger    cow
7  tiger    cow
8    cow  tiger

第二個數據集包含第一組項目的索引代碼：

index <- data.frame(item = c("cat", "dog", "tiger", "cow"),
                    code = c(1, 0, 1, 0))
index
   item code
1   cat    1
2   dog    0
3 tiger    1
4   cow    0

我正在尋找一種方法來創建兩個新列： tag0和tag1 ，使其看起來像這樣：

new <- data.frame(label1 = c("cat", "cat", "dog", "dog", "cat", "tiger", "tiger", "cow"),
                  label2 = c("dog", "dog", "cat", "cat", "dog", "cow", "cow", "tiger"),
                  tag1 = c("cat", "cat", "cat", "cat", "cat", "tiger", "tiger", "tiger"),
                  tag0 = c("dog", "dog", "dog", "dog", "dog", "cow", "cow", "cow"))
new
  label1 label2  tag1 tag0
1    cat    dog   cat  dog
2    cat    dog   cat  dog
3    dog    cat   cat  dog
4    dog    cat   cat  dog
5    cat    dog   cat  dog
6  tiger    cow tiger  cow
7  tiger    cow tiger  cow
8    cow  tiger tiger  cow

tag0是指label對應code=0 ， tag1指label對應index dataframe中code=1 。

任何人都可以通過基於tidyverse的解決方案幫助我嗎？

Answer 1

這是tidyverse中的兩個解決方案。 雖然第一個適用於這種特殊情況，但我更喜歡第二個，它更優雅且可擴展。

解決方案 1：每個label*的JOIN *

首先導入tidyverse並生成您的數據集original和index 。

library(tidyverse)


# ...
# Code to generate 'original' and 'index' datasets.
# ...

然后應用此工作流程。

original %>%
  # Uniquely identify each row (for pivoting later).
  mutate(row_id = row_number()) %>%
  # Match 'label1' to the tags.
  left_join(
    index,
    by = c("label1" = "item"),
    keep = TRUE
  ) %>%
  # Match 'label2' to the tags.
  left_join(
    index,
    by = c("label2" = "item"),
    keep = TRUE,
    suffix = c(".1", ".2")
  ) %>%
  # Pivot 'item.1 | ... | item.n | code.1 | ... | code.n' into a consolidated
  # 'item | code' form.
  pivot_longer(
    cols = matches("^(item|code)\\.(\\d+)?$"),
    names_pattern = "^(item|code)\\.(\\d+)?$",
    names_to = c(".value", NA)
  ) %>%
  # Pivot back into a 'tag1 | tag0' form.
  pivot_wider(
    values_from = item,
    names_from = code,
    names_glue = "tag{code}"
  ) %>%
  # Omit unique identifier.
  select(!row_id)

結果

鑒於此處復制的original數據集和index數據集

original <- data.frame(
  label1 = c("cat", "cat", "dog", "dog", "cat", "tiger", "tiger", "cow"),
  label2 = c("dog", "dog", "cat", "cat", "dog", "cow", "cow", "tiger")
)

index <- data.frame(
  item = c("cat", "dog", "tiger", "cow"),
  code = c(1, 0, 1, 0)
)

該解決方案應產生以下結果：

# A tibble: 8 x 4
  label1 label2 tag1  tag0 
  <chr>  <chr>  <chr> <chr>
1 cat    dog    cat   dog  
2 cat    dog    cat   dog  
3 dog    cat    cat   dog  
4 dog    cat    cat   dog  
5 cat    dog    cat   dog  
6 tiger  cow    tiger cow  
7 tiger  cow    tiger cow  
8 cow    tiger  tiger cow  

筆記

如果您的original數據集有任何其他label*列，則您需要為這些列中的每一列執行額外的JOIN 。

---

解決方案 2：單個CROSS JOIN

這是一個更優雅的工作流程，也更靈活：它適用於original中任意數量的label*列和index中的任意code集。

original %>%
  # Uniquely identify each row (for pivoting later).
  mutate(row_id = row_number()) %>%
  # Perform a cross-join compare every 'item' to every 'label*'.
  full_join(
    index,
    by = character()
  ) %>%
  # Keep only those rows where 'item' matches a 'label*'.
  rowwise() %>%
  filter(item %in% c_across(matches("^label\\d+"))) %>%
  # Pivot into a 'tag1 | tag0' form.
  pivot_wider(
    values_from = item,
    names_from = code,
    names_glue = "tag{code}"
  ) %>%
  # Omit unique identifier.
  select(!row_id)

結果

結果保持不變。

# A tibble: 8 x 4
  label1 label2 tag1  tag0 
  <chr>  <chr>  <chr> <chr>
1 cat    dog    cat   dog  
2 cat    dog    cat   dog  
3 dog    cat    cat   dog  
4 dog    cat    cat   dog  
5 cat    dog    cat   dog  
6 tiger  cow    tiger cow  
7 tiger  cow    tiger cow  
8 cow    tiger  tiger cow  

筆記

唯一的缺點是它必須執行CROSS JOIN ，這可能會阻礙更大數據集的性能。

Answer 2

另一種可能的解決方案：

library(tidyverse)

original <- data.frame(label1 = c("cat", "cat", "dog", "dog", "cat", "tiger", "tiger", "cow"),
                       label2 = c("dog", "dog", "cat", "cat", "dog", "cow", "cow", "tiger"))

index <- data.frame(item = c("cat", "dog", "tiger", "cow"),
                    code = c(1, 0, 1, 0))

original %>% 
  full_join(index, by=c("label1" = "item")) %>% 
  full_join(index, by=c("label2" = "item")) %>% 
  mutate(tag1 = if_else(code.x == 1, label1, label2)) %>% 
  mutate(tag2 = if_else(code.y == 1, label1, label2)) %>% 
  select(!starts_with("code"))

#>   label1 label2  tag1 tag2
#> 1    cat    dog   cat  dog
#> 2    cat    dog   cat  dog
#> 3    dog    cat   cat  dog
#> 4    dog    cat   cat  dog
#> 5    cat    dog   cat  dog
#> 6  tiger    cow tiger  cow
#> 7  tiger    cow tiger  cow
#> 8    cow  tiger tiger  cow