# [R] Generating a "conditional time" variable

Finak Greg Greg.Finak at ircm.qc.ca
Sat May 9 22:10:54 CEST 2009

```That will teach me to post without a double-check.

On 09/05/09 3:11 PM, "Finak Greg" <Greg.Finak at ircm.qc.ca> wrote:

Assuming the year column has complete data and doesn't skip a year, the following should take care of 1)

#Simulated data frame: year from 1990 to 2003, for 5 different ids, each having one or two eif "events"
test<-data.frame(year=rep(1990:2003,5),id=gl(5,length(1990:2003)),eif=as.vector(sapply(1:5,function(z){a<-rep(0,length(1990:2003));a[sample(1:length(1990:2003),sample(1:2,1))]<-1;a})))

#Generate the "conditional_time" column.
test<-do.call("rbind",lapply(split(test,test\$id),function(z){s<-0;data.frame(z,cond_time=sapply(z\$eif,function(i)ifelse(i==1,s<-1,s<<-s+1)))}))

The above resets the count at eif==1 rather than after, and there's a local assignment to s which should be global.
Thanks, David, for noting that.

do.call("rbind",lapply(split(test,test\$id),function(z){s<-0;data.frame(z,cond_time=sapply(z\$eif,function(i)ifelse(i==1,{l<-s+1;s<<-0;l},{l<-s+1;s<<-s+1;l})))}))

Generally sapply, lapply, and apply are faster than "for" loops. split() will split your data frame by the \$id column (second argument). lapply() loops through the resulting list and generates the cond_time variable, resetting when eif==1, otherwise incrementing the count, much as you have in your code.

If I understand 2) correctly, the following should do the trick:
test2<-test; #copy the data frame
test2<-do.call("rbind",lapply(split(test,test\$id),function(z)z[1:which(z\$eif==1)[1],]))

Similar to the former, but sub-setting the rows of the data data frame up to the first event, for each id.

If the above is all you need, then 1) and 2) could be combined in a single call.

Others will likely have a different approach..

Cheers,

--
Greg Finak
Post-Doctoral Research Associate
Computational Biology Unit
Institut des Recherches Cliniques de Montreal
Montreal, QC.

On 09/05/09 1:40 PM, "Vincent Arel-Bundock" <vincent.arel at gmail.com> wrote:

Hi everyone,

Please forgive me if my question is simple and my code terrible, I'm new to
it if someone could share conceptual hints for programming, or point me
toward an R function/package that could speed up my processing time.

Thanks a lot for your help!

##

My dataframe includes the variables 'year', 'id', and 'eif' and has +/- 1.9
million id-year observations

I would like to do 2 things:

-1- I want to create a 'conditional_time' variable, which increases in
increments of 1 every year, but which resets during year(t) if event 'eif'
occured for this 'id' at year(t-1). It should also reset when we switch to a
new 'id'. For example:

dataframe = test
year        id         eif  conditional_time

1990       1010          0    1
1991       1010          0    2
1992       1010          1    3
1993       1010          0    1
1994       1010          0    2
1995       1010          0    3
1996       1010          0    4
1997       1010          1    5
1998       1010          0    1
1999       1010          0    2
2000       1010          0    3
2001       1010          0    4
2002       1010          0    5
2003       1010          0    6
1990       2010          0    1
1991       2010          0    2
1992       2010          0    3
1993       2010          0    4
1994       2010          0    5
1995       2010          0    6
1996       2010          0    7
1997       2010          0    8
1998       2010          0    9
1999       2010          0    10
2000       2010          0    11
2001       2010          1    12
2002       2010          0    1
2003       2010          0    2

-2- In a copy of the original dataframe, drop all id-year rows that
correspond to years after a given id has experienced his first 'eif' event.

I have written the code below to take care of -1-, but it is incredibly
inefficient. Given the size of my database, and considering how slow my
computer is, I don't think it's practical to use it. Also, it depends on
correct sorting of the dataframe, which might generate errors.

##

for (i in 1:nrow(test)) {
if (i == 1) {                            # If first id-year
cond_time <- 1
test[i, 4] <- cond_time

} else if ((test[i-1, 1]) != (test[i, 4])) {             # If new id
cond_time <- 1
test[i, 4] <- cond_time
} else {                            # Same id as previous row
if (test[i, 3] == 0) {
test[i, 4] <- sum(cond_time, 1)
cond_time <- test[i, 6]
} else {
test[i, 4] <- sum(cond_time, 1)
cond_time <- 0
}
}
}

--
Vincent Arel
M.A. Student, McGill University

[[alternative HTML version deleted]]

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help