从 R 中的宽数据帧中检索连续事件的计数

Question

我正在使用 R 中的宽格式数据集。它是用户生成的数据集，每个 UserID 的每一行显示每个日期的连续事件。这是我正在使用的数据类型的模拟示例：-

df<-structure(list(UserID = c("hdyyu-192", "yeui-1893", "dnnd-1882", "nopr-738", "ieka-1728"), 
                   `05/06/2021` = c("Activity B", "Activity A", "Activity B", "Activity C", "Activity D"), 
                   `06/06/2021` = c("Activity B", "Activity A", "Activity A", "Activity C", "Activity D"), 
                   `07/06/2021` = c("Activity D","Activity A", NA, "Activity C", "Activity D"), 
                   `08/06/2021` = c("Activity A","Activity B", NA, "Activity B", "Activity A"), 
                   `09/06/2021` = c("Activity A","Activity B", NA, "Activity B", "Activity C"), 
                   `10/06/2021` = c("Activity C",NA, NA, NA, NA), 
                   `11/06/2021` = c("Activity B", NA, NA, NA, NA)), 
              class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA,-5L))


head(df)

# A tibble: 5 x 8
  UserID    `05/06/2021` `06/06/2021` `07/06/2021` `08/06/2021` `09/06/2021` `10/06/2021` `11/06/2021`
  <chr>     <chr>        <chr>        <chr>        <chr>        <chr>        <chr>        <chr>       
1 hdyyu-192 Activity B   Activity B   Activity D   Activity A   Activity A   Activity C   Activity B  
2 yeui-1893 Activity A   Activity A   Activity A   Activity B   Activity B   NA           NA          
3 dnnd-1882 Activity B   Activity A   NA           NA           NA           NA           NA          
4 nopr-738  Activity C   Activity C   Activity C   Activity B   Activity B   NA           NA          
5 ieka-1728 Activity D   Activity D   Activity D   Activity A   Activity C   NA           NA

我想做的是计算连续事件的频率；换句话说，antecedent/consequent 事件的频率。

这是所需的输出：-

#desired outcome

Antecedent   |   Consequent   |  Count

Activity A   |   Activity A   |    3
Activity A   |   Activity B   |    1
Activity A   |   Activity C   |    1

........................................

Activity D   |  Activity A   |    2
Activity D   |  Activity B   |    0
Activity D   |  Activity C   |    0
Activity D   |  Activity D   |    2

有人可以在 R 中建议一种可以让我实现此输出的方法吗？非常感谢:)

Answer 1

我会使用字符串搜索。我们可以使用 expand.grid 创建所有可能的组合，然后使用 do.call 将它们转换为字符串，在 df 上做同样的事情（没有 UserID 列）并且只需使用 stringr::str_count() 获取每一行中的实例数，总和是它在 df 中出现的总次数。这有助于我们维持您想要的顺序。

注意，我们需要在正则表达式中使用前瞻来计算重叠实例。

library(stringr)

activities <- paste0("Activity ", LETTERS[1:5])
activities_df <- expand.grid(activities, activities)

activities_df$count <- sapply(
  do.call(paste, activities_df),
  \(x) sum(str_count(do.call(paste, df[,-1]), paste0("(?=", x, ")")))
)

head(activities_df)
#>         Var1       Var2 count
#> 1 Activity A Activity A     3
#> 2 Activity B Activity A     1
#> 3 Activity C Activity A     0
#> 4 Activity D Activity A     2
#> 5 Activity E Activity A     0
#> 6 Activity A Activity B     1

_{注意\(x)是R >= 4.1.0中function(x)的shorthand，如果使用以前的版本就改成function(x) .}

Answer 2

另一种可能的解决方案，使用tidyverse：

library(tidyverse)

df %>% 
  pivot_longer(-UserID) %>% 
  group_by(UserID) %>% 
  mutate(aux = str_c(lag(value), value, sep = "-")) %>% 
  ungroup %>% select(aux) %>% 
  mutate(aux2 = "count") %>% 
  pivot_wider(id_cols = aux, names_from = aux2, 
     values_from = aux2, values_fn = length) %>% drop_na(aux) %>%
  separate(aux, into = c("Antecedent", "Consequent"), sep = "-") %>% 
  arrange(Antecedent, Consequent)

#> # A tibble: 10 × 3
#>    Antecedent Consequent count
#>    <chr>      <chr>      <int>
#>  1 Activity A Activity A     3
#>  2 Activity A Activity B     1
#>  3 Activity A Activity C     2
#>  4 Activity B Activity A     1
#>  5 Activity B Activity B     3
#>  6 Activity B Activity D     1
#>  7 Activity C Activity B     2
#>  8 Activity C Activity C     2
#>  9 Activity D Activity A     2
#> 10 Activity D Activity D     2

如果有人有兴趣也显示零计数：

df %>% 
  pivot_longer(-UserID) %>% 
  group_by(UserID) %>% 
  mutate(aux = str_c(lag(value), value, sep = "-")) %>% 
  ungroup %>% select(aux) %>% 
  mutate(aux2 = "count") %>% 
  pivot_wider(id_cols = aux, names_from = aux2, 
     values_from = aux2, values_fn = length) %>% drop_na(aux) %>% 
  separate(aux, into = c("Antecedent", "Consequent"), sep = "-") %>% 
  bind_rows(df[-1] %>% unlist %>% as.vector() %>% unique %>% 
              expand_grid(Antecedent = ., Consequent = .) ) %>% 
  drop_na(c(Antecedent, Consequent)) %>% 
  group_by(Antecedent, Consequent) %>% 
  summarise(count = sum(count, na.rm = T), .groups = "drop") %>% 
  arrange(Antecedent, Consequent)

#> # A tibble: 16 × 3
#>    Antecedent Consequent count
#>    <chr>      <chr>      <int>
#>  1 Activity A Activity A     3
#>  2 Activity A Activity B     1
#>  3 Activity A Activity C     2
#>  4 Activity A Activity D     0
#>  5 Activity B Activity A     1
#>  6 Activity B Activity B     3
#>  7 Activity B Activity C     0
#>  8 Activity B Activity D     1
#>  9 Activity C Activity A     0
#> 10 Activity C Activity B     2
#> 11 Activity C Activity C     2
#> 12 Activity C Activity D     0
#> 13 Activity D Activity A     2
#> 14 Activity D Activity B     0
#> 15 Activity D Activity C     0
#> 16 Activity D Activity D     2

Answer 3

与data.table:

library(data.table)
setDT(df)
DT <- melt(df,id.vars = "UserID",variable.name = 'Timestamp',value.name = 'Activity')
DT[,TimeStamp:=as.Date(Timestamp,format='%d/%m/%Y')]
DT <- DT[order(Timestamp)][,Activity_prec:=shift(Activity),by=.(UserID)]
DT[!is.na(Activity_prec)&!is.na(Activity),.(Count=.N),by=.(Activity_prec,Activity)][
   order(Activity_prec,Activity)]

    Activity_prec   Activity Count
           <char>     <char> <int>
 1:    Activity A Activity A     3
 2:    Activity A Activity B     1
 3:    Activity A Activity C     2
 4:    Activity B Activity A     1
 5:    Activity B Activity B     3
 6:    Activity B Activity D     1
 7:    Activity C Activity B     2
 8:    Activity C Activity C     2
 9:    Activity D Activity A     2
10:    Activity D Activity D     2

从 R 中的宽数据帧中检索连续事件的计数

Retrieving counts of consecutive events from wide dataframe in R

r

dplyr

data.table