根据其他列与df中其他行的关系更改一行中的df列值

Question

Working in R 3.1.1.

I have a dataset with transaction data. Each customer has bought at least twice (I've subseted my original data). What I would like to do, is flag each transaction as a "first time buyer" transaction or "repeat buyer" transaction. The issue is, I would like to define a "repeat buyer transaction" as one within a certain time frame of a past transaction, so it's not quite as simple as flagging the first one ever for each customer as "first" and the rest as "repeat". If a customer hasn't bought in more than 1 year (52.25 weeks, I want him/her to be counted as first time!)

The best way of accomplishing this that I've been able to come up with is extremely inefficient, I think (full disclosure, still running, so it may be erroneous to boot). I'm using nested for loops... :(

Any suggestions on how to accomplish this more efficiently? Thanks in advance for your help and suggestions! Code is commented throughout so I'll let it speak for itself, but please do let me know if it's not clear!

 #let's ensure the repdata is ordered by date first attach(repdata) repdata <- repdata[order(date),] detach(repdata) #now, we loop through repdata and decide whether purchase #is a first time or repeat buyer #setting time frame to 1 year (52.25 weeks as we use week as units below) timeframe = 52.25 #add new column to repdata that we will use below repdata$rpt52wk <- "" #for each row in repdata, do the following for(i in seq_along(repdata$date)) { #assume that this is a first purchase; set rpt52wk var for [i] to "FIRST TIME BUYER" repdata$rpt52wk[i] = "FIRST TIME BUYER" #look at all previous transactions #we can ignore higher indexed transactions (we sorted the data, ascending by date) for (j in seq_along(repdata$date[1:(i-1)])) { #if a transaction is found in which the same member bought within the timeframe else if(repdata$MEMBER_ID[i] == repdata$MEMBER_ID[j] & (difftime(repdata$date[i],repdata$date[j],units="weeks")<timeframe)) { #then this is a repeat buyer; set rpt var for [i] appropriately repdata$rpt52wk[i]="REPEAT BUYER" } } } 

Adding test data that fails, at least when run on my side with the two solutions presented so far.

MEMBER_ID       date
      1 2011-04-13
      2 2011-04-22
      3 2011-04-17
      3 2011-04-26
      4 2011-04-13
      4 2011-04-16
      4 2011-04-16
      5 2011-04-20
      5 2011-04-13
      5 2011-04-18
      6 2011-04-13
      7 2011-04-13
      8 2011-04-25
      8 2011-04-20
      9 2011-04-14
     10 2011-04-14
     11 2011-04-18
     12 2011-04-15
     13 2011-04-15
     14 2011-04-13

#TEST SET GENERATION:
library(lubridate)
MEMBER_ID <- c(1,2,3,3,4,4,4,5,5,5,6,7,8,8,9,10,11,12,13,14)
date <- ymd(c("2011-04-13 UTC", "2011-04-22 UTC", "2011-04-17 UTC", "2011-04-26 UTC", 
          "2011-04-13 UTC", "2011-04-16 UTC", "2011-04-16 UTC", "2011-04-20 UTC", 
          "2011-04-13 UTC", "2011-04-18 UTC", "2011-04-13 UTC", "2011-04-13 UTC", 
          "2011-04-25 UTC", "2011-04-20 UTC", "2011-04-14 UTC", "2011-04-14 UTC", 
          "2011-04-18 UTC", "2011-04-15 UTC", "2011-04-15 UTC", "2011-04-13 UTC"))
rm(repdata)
repdata <- data.frame(MEMBER_ID, date)
repdata

(Note that I realize that the code has a bug for i=1. I'm going to ignore it for now in favour of not adding another if statement inside my for loop)

Answer 1

You could give it a try using ddply.

First generate a dataset sorted by date, with a timeframe of 52 weeks.

#TEST SET GENERATION:
library(lubridate)
MEMBER_ID <- c(1,2,3,3,4,4,4,5,5,5,6,7,8,8,9,10,11,12,13,14)
date <- ymd(c("2011-04-13 UTC", "2011-04-22 UTC", "2011-04-17 UTC", "2011-04-26 UTC", 
          "2011-04-13 UTC", "2011-04-16 UTC", "2011-04-16 UTC", "2011-04-20 UTC", 
          "2011-04-13 UTC", "2011-04-18 UTC", "2011-04-13 UTC", "2011-04-13 UTC", 
          "2011-04-25 UTC", "2011-04-20 UTC", "2011-04-14 UTC", "2011-04-14 UTC", 
          "2011-04-18 UTC", "2011-04-15 UTC", "2011-04-15 UTC", "2011-04-13 UTC"))
rm(repdata)
repdata <- data.frame(MEMBER_ID, date)
repdata <- repdata[order(repdata$date),]
repdata

# define a timeframe of 4 weeks
timeframe <- as.difftime(52, units = "weeks")

Then tun the following code :

library(plyr)

first.buyers <- ddply(repdata, .(MEMBER_ID),
                  function(x) x[c(TRUE, diff(x$date) > timeframe),])
first.buyers <- mutate(first.buyers, rpt52wk = "FIRST TIME BUYER")

final <- merge(repdata,first.buyers, all = TRUE)
final[is.na(final$rpt52wk),"rpt52wk"] <- "REPEAT BUYER"

We get the following result :

   MEMBER_ID       date          rpt52wk
1          1 2011-04-13 FIRST TIME BUYER
2          2 2011-04-22 FIRST TIME BUYER
3          3 2011-04-17 FIRST TIME BUYER
4          3 2011-04-26     REPEAT BUYER
5          4 2011-04-13 FIRST TIME BUYER
6          4 2011-04-16     REPEAT BUYER
7          4 2011-04-16     REPEAT BUYER
8          5 2011-04-13 FIRST TIME BUYER
9          5 2011-04-18     REPEAT BUYER
10         5 2011-04-20     REPEAT BUYER
11         6 2011-04-13 FIRST TIME BUYER
12         7 2011-04-13 FIRST TIME BUYER
13         8 2011-04-20 FIRST TIME BUYER
14         8 2011-04-25     REPEAT BUYER
15         9 2011-04-14 FIRST TIME BUYER
16        10 2011-04-14 FIRST TIME BUYER
17        11 2011-04-18 FIRST TIME BUYER
18        12 2011-04-15 FIRST TIME BUYER
19        13 2011-04-15 FIRST TIME BUYER
20        14 2011-04-13 FIRST TIME BUYER

ddply splits your dataframe by MEMBER_ID, and apply a function to each subset. Each subset is a dataframe with fixed MEMBER_ID and ordered date. The first element will always correspond to a first buyer, for the next elements you have to determine if the time elapsed since the last transaction is larger than your threshold (If yes, this member can be again consider as a first buyer).

In the code above you should check that time units are consistent when doing the comparison diff(x$date) > timeframe (depends on your date format)

Once you have found the first time buyers I think the next steps are rather explicit.