Aggregating Data in R with user defined function

Question

I have grouped data in R using the aggregate method.

Avg=aggregate(x$a, by=list(x$b,x$c),FUN= mean)

This gives me the mean for all the values of 'a' grouped by 'b' and 'c' of data frame 'x'.

Now instead of taking the average of all values of 'a' I want to take the average of 3 maximum values of 'a' grouped by 'b' and 'c' .

Sample data set

a    b    c
10   G    3 
20   G    3 
22   G    3
10   G    3 
15   G    3
25   G    3
30   G    3

After above Aggregate function it will give me

Group.1    Group.2    x
  G          3       18.85

But I want to take just the maximum 5 values of 'a' for average

Group.1    Group.2    x
  G          3       22.40

I am not able to accommodate the below maximum function that i am using in the Agrregate function

index <- order(vector, decreasing = T)[1:5]
vector(index)

Can please anyone throw some light on how is this possible ?

Answer 1

You can order the data, get the top 5 entries (using head) and then apply the mean:

aggregate(x$a, by=list(x$b,x$c),FUN= function(x) mean(head(x[order(-x)], 5)))
#  Group.1 Group.2    x
#1       G       3 22.4

If you want to do this with a custom function, I would do it like this:

myfunc <- function(vec, n){
  mean(head(vec[order(-vec)], n))
}

aggregate(x$a, by=list(x$b,x$c),FUN= function(z) myfunc(z, 5))
#  Group.1 Group.2    x
#1       G       3 22.4

I actually prefer using the formula style in aggregate which would look like this (I also use with() to be able to refer to the column names directly without using x$ each time):

with(x, aggregate(a ~ b + c, FUN= function(z) myfunc(z, 5)))
#  b c    a
#1 G 3 22.4

In this function, the parameter z is passed each a-vector based on groups of b and c. Does that make more sense now? Also note that it doesn't return an integer here but a numeric (decimal, 22.4 in this case) value.