因子水平出现的计数顺序

Question

My transactional dataset contains several variables including an ID number, date, and transaction area (factor):

    id<-as.integer(rep(c(1,2,3,4,5),times=20))
    date<-rep(seq(as.Date("2011-07-01"),by="day", length.out=100))
    category<-rep(as.factor(letters[seq( from = 1, to = 4 )]),times=25)

    transactions<-data.frame(id, date, category)

    head(transactions)
      id       date category
       1 2011-07-01        a
       2 2011-07-02        b
       3 2011-07-03        c
       4 2011-07-04        d
       5 2011-07-05        a
       1 2011-07-06        b

What I would like to do on a per-ID basis is determine the order of the factor appearance without recounting a factor that has already appeared.

    solution <- transactions %>%
                group_by(id, date)%>%
                mutate(category_order= (solution))

So I can get:

     head(transactions)
      id       date category category_order
       1 2011-07-01        a     1
       1 2011-07-06        b     2
       1 2011-07-11        c     3
       1 2011-07-16        d     4
       1 2011-07-21        a     1
       1 2011-07-26        b     2

For each ID if a category repeats, it must have the same order value. In the example above a is always 1st, b is always 2nd, etc.

What I want to do is count the number of times each category is 1st, 2nd, 3rd, etc. to obtain a frequency distribution for number of times a is 1st, b is 1st, a is 2nd, etc:

    head(transactions)
       category category_ order category_order_count
       a     1     5
       a     2     3
       a     3     5
       a     4     4
       b     1     5
       b     2     2

It's probably not complicated, but I am having a mental block because it essentially involves counting an order per ID without repeating a factor level, then summarizing per ID, and finally summarizing per category.

Answer 1

Within each id you could set the levels of the factor to the order they appear within that group and then transform the factors to integers via as.numeric to form your new variable. This relies on the order of the dataset, so if things aren't in order you should arrange by id and date.

transactions %>%
    arrange(id, date) %>%
    group_by(id) %>%
    mutate(category_order = as.numeric(factor(category, levels = unique(category))))

Answer 2

This can be also done with data.table . We convert the 'data.frame' to 'data.table' ( setDT(transactions) ), grouped by 'id', specify the 'i' part with order of 'id' and 'date', convert the 'category' into factor by specifying the levels as the unique elements in 'category', coerce it to 'integer' and assign ( := ) the output to 'category_order'

library(data.table)
setDT(transactions)[order(id, date), category_order := as.integer(factor(category,
            levels = unique(category))) , by = id]

head(transactions)
#   id       date category category_order
#1:  1 2011-07-01        a              1
#2:  2 2011-07-02        b              1
#3:  3 2011-07-03        c              1
#4:  4 2011-07-04        d              1
#5:  5 2011-07-05        a              1
#6:  1 2011-07-06        b              2