R - ddply

ddply

dataframe in dataframe out

example

• first we create a data frame with the following columns: id, date, value
• to create the id column we use rep(), where the first argument 1:5 (the vector (1,2,3,4,5)) is repeated five times, so we have a vector with length 25
• the vector dates is generated in the following way:
  • generate a vector of random numbers between 1 and 1500 and length 25 (sample.int())
  • use as.Date with the origin argument, this converts the integers into Date, where the the origin is day 0 (in the case 0 is convert to the origin i.e. 2000-01-01, 1 is mapped to 2000-01-02 etc pp, so we get a vector of dates between 2000 and beginning 2005
• vals are generated by the random numbers function rnorm() (normal distributed with mean 5 and standard deviation 1
• last but not least put the three together in one data frame (and let us show the first lines)

id <- rep(1:5,5)
dates <- as.Date(sample.int(1500, 25, replace=T), origin="2000-01-01") 
vals <- rnorm(25, mean=5, sd=1)
df <- data.frame(id=id, date=dates, val=vals)
head(df)

  id       date      val
1  1 2001-12-31 5.778680
2  2 2002-08-02 6.982799
3  3 2002-04-23 5.925903
4  4 2000-08-03 3.527375
5  5 2002-08-29 5.239211
6  1 2003-01-28 5.118337

• id encodes a person, date contains the day of the measurement and val the values
• now we want to add a column which contains the the first measurement in time
• therefore we had to load plyr, then we use the function ddply() (the meaning of the first two letters is data frame in data frame out)
• the first argument of the function is the data frame we pass to ddply
• the second argument defines the groups (we want the min of each person so our grouping variable is id)
• transform says we want to change existing data frame - like recode a variable or add a new (another choice would be summarise - if we want to aggregate the data) - so we add a variable named start and it should be the min() of our date per person

res <- ddply(df, .(id), transform, start=min(date))
res

id	date	val	start
1	2001-12-31	5.77867959202519	2001-03-07
1	2003-01-28	5.11833655328897	2001-03-07
1	2003-11-14	3.75075463437505	2001-03-07
1	2001-03-07	3.36305093773468	2001-03-07
1	2001-03-07	6.6141583789233	2001-03-07
2	2002-08-02	6.98279884579167	2000-05-29
2	2003-12-30	5.4450019138646	2000-05-29
2	2000-05-29	5.92567982716667	2000-05-29
2	2000-11-26	4.60169956597544	2000-05-29
2	2002-08-04	5.02395812458	2000-05-29
3	2002-04-23	5.92590324593659	2000-02-02
3	2003-04-28	6.59246056959167	2000-02-02
3	2002-10-20	5.89780464300662	2000-02-02
3	2000-02-02	4.0548618816269	2000-02-02
3	2000-05-30	5.4722633402378	2000-02-02
4	2000-08-03	3.52737458048037	2000-08-03
4	2002-11-02	3.98853894162285	2000-08-03
4	2002-11-13	4.54373088519655	2000-08-03
4	2003-08-02	4.2144245184346	2000-08-03
4	2003-12-19	4.98767478298279	2000-08-03
5	2002-08-29	5.23921121209821	2000-04-27
5	2000-04-27	4.20400241411961	2000-04-27
5	2003-07-26	6.1821957272646	2000-04-27
5	2003-05-18	4.53515404638362	2000-04-27
5	2001-07-23	3.06695274940501	2000-04-27

• if we change the function just a little we get the days elapsed from the first measurment:

res <- ddply(df, .(id), transform, dayselapsed=date-min(date))
res

id	date	val	dayselapsed
1	2001-12-31	5.77867959202519	299
1	2003-01-28	5.11833655328897	692
1	2003-11-14	3.75075463437505	982
1	2001-03-07	3.36305093773468	0
1	2001-03-07	6.6141583789233	0
2	2002-08-02	6.98279884579167	795
2	2003-12-30	5.4450019138646	1310
2	2000-05-29	5.92567982716667	0
2	2000-11-26	4.60169956597544	181
2	2002-08-04	5.02395812458	797
3	2002-04-23	5.92590324593659	811
3	2003-04-28	6.59246056959167	1181
3	2002-10-20	5.89780464300662	991
3	2000-02-02	4.0548618816269	0
3	2000-05-30	5.4722633402378	118
4	2000-08-03	3.52737458048037	0
4	2002-11-02	3.98853894162285	821
4	2002-11-13	4.54373088519655	832
4	2003-08-02	4.2144245184346	1094
4	2003-12-19	4.98767478298279	1233
5	2002-08-29	5.23921121209821	854
5	2000-04-27	4.20400241411961	0
5	2003-07-26	6.1821957272646	1185
5	2003-05-18	4.53515404638362	1116
5	2001-07-23	3.06695274940501	452