How to speed up cummulative sum within group?

问题

I have the following data frame:

id<-c(1,1,1,1,1,3,3,3,3)
spent<-c(10,20,30,40,50,60,70,80,90)
date<-c("11-11-07","11-11-07","23-11-07","12-12-08","17-12-08","11-11-07","23-11-07","23-   11-07","16-01-08")
df<-data.frame(id,date,spent)
df$date2<-as.Date(as.character(df$date), format = "%d-%m-%y")


 id     date spent      date2
1  1 11-11-07    10 2007-11-11
2  1 11-11-07    20 2007-11-11
3  1 23-11-07    30 2007-11-23
4  1 12-12-08    40 2008-12-12
5  1 17-12-08    50 2008-12-17
6  3 11-11-07    60 2007-11-11
7  3 23-11-07    70 2007-11-23
8  3 23-11-07    80 2007-11-23
9  3 16-01-08    90 2008-01-16

I need to calculate the sum spent by each id per day and include it in the frame work as follow:

 id     date spent      date2    sum.spent
1  1 11-11-07    10 2007-11-11    10
2  1 11-11-07    20 2007-11-11    30 
3  1 23-11-07    30 2007-11-23    30
4  1 12-12-08    40 2008-12-12    40
5  1 17-12-08    50 2008-12-17    50
6  3 11-11-07    60 2007-11-11    60
7  3 23-11-07    70 2007-11-23    70
8  3 23-11-07    80 2007-11-23    150
9  3 16-01-08    90 2008-01-16    90

The following script works well (except for the first row which is not a big deal):

df$spent2<-NA
for (a in 2:9)
if (df[a,1]==df[a-1,1]&& df[a,4]==df[a-1,4])
(df[a,5]=df[a,3]+df[a-1,3])else(df[a,5]=df[a,3])

However since the number of rows in my actual dataset is around 1.5 million, the above script takes around 5 days to be executed. I wonder if you can suggest a more efficient way to write this code and achieve the same objective.

回答1:

data.table is pretty fast, especially for such large datasets. This should run pretty quickly for 1.5 mil records.

library(data.table)
df <- data.table(df)
df <- df[, sum.spent:=cumsum(spent), by = list(id, date2)]

回答2:

Here is a base R solution:

df$sum.spent <- ave(df$spent,df$id,df$date2,FUN=cumsum)

I get a different result than your expected answer though, but I think it is correct?

来源：https://stackoverflow.com/questions/13081821/how-to-speed-up-cummulative-sum-within-group

标签

aggregate