Conditional accumulating sum with a dynamic condition - r

Question

Good afternoon
i'm trying to create an accumulating mean with a "twist" - i only want to average fields that dated before the current one (there may be field with the same date)
i succeeded doing it the "dirty way" using several custom created function but it takes too long and it's very inefficient - im pretty sure that there is a better way.
i was thinking about something along the line of:
averages <- DB %>% group_by(field1,field2) %>% mutate(Avg=cummean(???*value1)))
how do i access the current observation for cummean function
the route i went was to create a logical vector for every subset with a loop
for (i in 1:length(datevector)-1)
logicalvector[i] <- datevector[length(datevector)]>datevector[i]
logicalvector[length(datevector)]=F
and use that in another function to calculate the mean
a simple example is:
df <- data.frame(id=1:5,Date=as.Date(c("2013-08-02","2013-08-02","2013-08-03","2013-08-03","2013-08-04")),Value=c(1,4,5,2,4))
id Date Value accum mean
1 02/08/2013 1 0
2 02/08/2013 4 0
3 03/08/2013 5 2.5
4 03/08/2013 2 2.5
5 04/08/2013 4 3
Explanation:
there are no observation with a prior date for the first 2 observations so the mean is 0
the 3rd observation averages the 1st and 2nd, so does the 4th.
the 5th observation averages all

Answer 1

This can be implwemented as a complex self-join in SQL. This joins to each row all rows with lesser Date amd for each row averages Value in the joined rows. coalesce is used to assign 0 in the situation where the mean would otherwise be Null.
library(sqldf)
sqldf("select a.*, coalesce(avg(b.Value), 0) as mean
from df as a
left join df as b on b.Date < a.Date
group by a.rowid")
giving:
id Date Value mean
1 1 2013-08-02 1 0.0
2 2 2013-08-02 4 0.0
3 3 2013-08-03 5 2.5
4 4 2013-08-03 2 2.5
5 5 2013-08-04 4 3.0

Answer 2

Using data.table and lubridate you have this option:
library(data.table)
library(lubridate)
dt <- data.table(id=c(1:5))
dt$Date <- c("02/08/2013", "02/08/2013", "03/08/2013", "03/08/2013", "04/08/2013")
dt$Value <- c(1,4,5,2,4)
dt$Date <- dmy(dt$Date)
cummean <- function(d){
if(nrow(dt[Date<d])>0)
dt[Date<d, sum(Value)/.N]
else 0
}
dt[, accuMean:=mapply(cummean,Date)]
# id Date Value accuMean
#1: 1 2013-08-02 1 0.0
#2: 2 2013-08-02 4 0.0
#3: 3 2013-08-03 5 2.5
#4: 4 2013-08-03 2 2.5
#5: 5 2013-08-04 4 3.0
Solution when you have multiple values:
library(data.table)
library(lubridate)
dt <- data.table(id=c(1:5))
dt$Date <- c("02/08/2013", "02/08/2013", "03/08/2013", "03/08/2013", "04/08/2013")
dt$Value_1 <- c(1,4,5,2,4)
dt$Value_2 <- c(3,2,0,1,2)
dt$Value_3 <- c(4,9,3,3,3)
dt$Date <- dmy(dt$Date)
cummean <- function(d,Value){
if(nrow(dt[Date<d])>0)
sum(dt[Date<d, Value, with=F])/dt[Date<d, .N]
else 0
}
n <- 3
accuMean <- paste0("accuMean_", (1:n))
for(i in 1:n){
print(i)
dt[, (accuMean[i]):=mapply(cummean,Date,MoreArgs = list(paste0("Value_",i)))]
}
Suppose you have n values named Value_i. Ten in your case, you only need to set n=10