R - Taking column maxes over specific subsets of a dataframe - r

Question

I have a large set of data with various indices and such. I would like to change my data from something like this:
id time var1_t1 var1_t2 var1_t3 var2_t1 var2_t2 var2_t3
1 1 1.5 NA NA 3.3 NA NA
1 2 NA 2.5 NA NA 1.2 NA
1 3 NA NA 3.5 NA NA .7
to something like this:
id time var1_t1 var1_t2 var1_t3 var2_t1 var2_t2 var2_t3
1 1 1.5 2.5 3.5 3.3 1.2 .7
1 2 NA 2.5 NA NA 1.2 NA
1 3 NA NA 3.5 NA NA .7
or this:
id time var1_t1 var1_t2 var1_t3 var2_t1 var2_t2 var2_t3
1 1 1.5 2.5 3.5 3.3 1.2 .7
1 2 1.5 2.5 3.5 3.3 1.2 .7
1 3 1.5 2.5 3.5 3.3 1.2 .7
Except that there are rather more than six columns I need to do this for, and "id' has values other than 1.
I can do this for a single column as follows:
for (i in 1:max(df$id) ){
df[df$id == i & df$time == 1,]$var1_t1 <- max(df[df$id == i,]$var1_t1,
na.rm = TRUE)
}
But that uses a for loop, so it is a terrible idea. And I would have to repeat that line for each column. Is there a way I can do this more elegantly?

Answer 1

If you want to replace all NA's with the column-wise max value by group of id, you could define a little custom function:
f <- function(x) {
x[is.na(x)] <- max(x, na.rm = TRUE)
x
}
And then use your favorite data manipulation functions/package, for example dplyr:
library(dplyr)
df %>% group_by(id) %>% mutate_each(funs(f))
Or data.table:
library(data.table)
setDT(df)[, lapply(.SD, f), by = id]