R对比列表识别逐月变化 unnest空列表丢行解决方案

unnest_longer()默认会丢弃列表列长度为0的对应行。你的代码中2022-03-01对应的flavors.removed是长度为0的空字符向量，因此整行被过滤，直接导致该月新增的Pumpkin口味记录丢失。

在调用unnest_longer()时传入keep_empty = TRUE参数，即可保留空列表对应的行，自动填充NA值；之后过滤掉非首月无移除口味时生成的冗余NA行，就能得到预期的9行完整结果。

完整可运行代码：

library(dplyr)
library(tidyr)

df %>% 
  group_by(date) %>% 
  summarize(flavors = list(flavor), .groups = "drop") %>% 
  mutate(
    flavors.added = mapply(setdiff, flavors, lag(flavors)),
    flavors.removed = mapply(setdiff, lag(flavors), flavors)
  ) %>% 
  select(-flavors) %>% 
  unnest_longer(flavors.added, keep_empty = TRUE) %>% 
  unnest_longer(flavors.removed, keep_empty = TRUE) %>% 
  pivot_longer(-date, names_to = "type", values_to = "flavor") %>% 
  # 过滤非首月无移除口味的空记录+去重
  filter(!(type == "flavors.removed" & is.na(flavor) & date != min(date))) %>% 
  unique() %>% 
  arrange(date, type)

运行结果与预期完全匹配：

# A tibble: 9 x 3
  date       type            flavor 
  <date>     <chr>           <chr>  
1 2022-01-01 flavors.added   Almond 
2 2022-01-01 flavors.added   Apple  
3 2022-01-01 flavors.added   Apricot
4 2022-01-01 flavors.removed NA     
5 2022-02-01 flavors.added   Maple  
6 2022-02-01 flavors.added   Mint   
7 2022-02-01 flavors.removed Apple  
8 2022-02-01 flavors.removed Apricot
9 2022-03-01 flavors.added   Pumpkin

如果不想额外写过滤逻辑，也可以提前对首月的空值、后续月份的空列表做统一替换，从根源避免长度为0的向量出现：

df %>% 
  group_by(date) %>% 
  summarize(flavors = list(flavor), .groups = "drop") %>% 
  mutate(
    prev_flavors = lag(flavors),
    flavors.added = Map(setdiff, flavors, prev_flavors),
    # 首月lag结果为NULL、无移除口味时直接返回NA，避免空列表
    flavors.removed = Map(\(curr, prev) {
      if (is.null(prev)) return(NA_character_)
      res <- setdiff(prev, curr)
      if (length(res) == 0) return(NA_character_)
      res
    }, flavors, prev_flavors)
  ) %>% 
  select(-flavors, -prev_flavors) %>% 
  unnest_longer(flavors.added) %>% 
  unnest_longer(flavors.removed) %>% 
  pivot_longer(-date, names_to = "type", values_to = "flavor") %>% 
  filter(!(is.na(flavor) & type == "flavors.added")) %>% 
  unique() %>% 
  arrange(date, type)

内容的提问来源于stack exchange，提问作者jophuh