Question

我有以下格式的交易数据。

transaction <- data.frame(account_id=c('ID001','ID001','ID002','ID002','ID001','ID002'), transaction_date=c('2017-01-02','2017-01-03','2017-01-03','2017-01-05','2017-01-06','2017-01-08'))

我想标记哪个帐户在2017年1月1日和2017年12月31日连续三个月处于非活动状态。这是我的R代码。

library(dplyr)

list_account_id <- transaction %>% distinct(account_id)
list_account_id <- list_account_id$account_id

churn_label <- data.frame("account_id" = c('ID000'), "churn_date" = c(as.Date('9999/99/99')), "label"=c(0))

for (int_account_id in list_account_id){
    start_date <- as.Date('2017-01-01')
    break_while <- FALSE
    while(!break_while){
        end_date <- start_date+90
        int_transaction <- transaction %>% filter(account_id == int_account_id)
        int_transaction %>% filter(as.Date(transaction_date) <= end_date, as.Date(transaction_date) >= start_date) %>% summarise(n=n())

        sum_ntransaction <- int_transaction %>% filter(as.Date(transaction_date) <= end_date, as.Date(transaction_date) >= start_date) %>% summarise(n=n())
        if(sum_ntransaction$n == 0){
            churn_label_temp <- data.frame("account_id" = c(int_account_id), "churn_date" = c(start_date), "label"=c(1))
            churn_label <- rbind(churn_label, churn_label_temp)
            break_while <- TRUE
        }
        if(end_date == as.Date('2017-12-31')){
            churn_label_temp <- data.frame("account_id" = c(int_account_id), "churn_date" = c('9999/99/99'), "label"=c(0))
            churn_label <- rbind(churn_label, churn_label_temp)
            break_while <- TRUE
        }
    start_date <- start_date+1
    }
}

我的代码的输出是

但是，我的代码太慢了。还有另一种方法吗？

Answer 1

tidyverse解决方案

还要考虑90天！= 3个月！对于几个月的计算，我更喜欢lubridate包中的%m+%运算符。

transaction <- data.frame(account_id=c('ID001','ID001','ID002','ID002','ID001','ID002'), transaction_date=c('2017-01-02','2017-01-03','2017-01-03','2017-01-05','2017-01-06','2017-01-08'))

library(tidyverse)
library(lubridate)

transaction %>%
  #set transaction date as dates
  mutate( transaction_date = as.Date( transaction_date ) ) %>%
  #group by account id
  group_by( account_id ) %>%
  #arrange on date 
  arrange( transaction_date ) %>%
  #inactive more than 3 months? Check if the next transaction date is larger than the currect transaction date + 3 months
  mutate( inactive_label = ifelse( transaction_date %m+% months(3) > lead( transaction_date ), 0, 1 ) ) %>%
  #also check the first and last row of each group (first not after 2017-01-01 + 3 months, last not before 31-12-2017 - 3 months)
  mutate( inactive_label = ifelse( row_number() == 1 & transaction_date > as.Date("2017-01-01") %m+% months(3), 1, inactive_label ) ) %>%
  mutate( inactive_label = ifelse( row_number() == n() & transaction_date %m+% months(3) < as.Date("2017-12-31") , 1, inactive_label ) )

# # A tibble: 6 x 3
# # Groups:   account_id [2]
# account_id transaction_date inactive_label
#   <fct>      <date>                    <dbl>
# 1 ID001      2017-01-02                    0
# 2 ID001      2017-01-03                    0
# 3 ID002      2017-01-03                    0
# 4 ID002      2017-01-05                    0
# 5 ID001      2017-01-06                    1
# 6 ID002      2017-01-08                    1

Answer 2

根据准确的客户流失定义，您可能只对最大交易日期感兴趣，并检查此交易距离受抚养人期限结束是否超过3个月。例如，您可以使用以下代码：

#Set up you timeline
end_dependent <- as.Date("2017-31-12", format= "%Y-%d-%m")
library(lubridate)
churndate <- end_dependent %m-% months(3)

# transaction date in date format
transaction$transaction_date <- as.Date(transaction$transaction_date,format= "%Y-%d-%m")

# Aggregate per account taking the maximum transaction date
transaction_per_account <- aggregate(list(max_transaction_date = transaction$transaction_date), by=list(account_id =transaction$account_id), max)
# Calculate a binary dependent variable for churn and churn date
transaction_per_account$churn <- ifelse(transaction_per_account$max_transaction_date <= churndate, 1, 0 )
ifelse(transaction_per_account$churn == 1, transaction_per_account$churn_date <- transaction_per_account$max_transaction_date, transaction_per_account$churn_date <- as.Date("9999-01-01", format= "%Y-%d-%m") )

这给出了输出：

> transaction_per_account
account_id max_transaction_date churn churn_date
1      ID001           2017-06-01     1 2017-06-01
2      ID002           2017-08-01     1 2017-08-01

R

2 个答案:

tidyverse解决方案