5

Sorry for the simple question but I can't think of a good way to take functions elements of a list of data frames. I am sure there is something within the plyr/reshape2 packages but I just can't think of it.

For example I have a list A as follows:

>A
[[1]]
        [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
   [1,]    1    1    1    1    1    1    1    1    1     1
   [2,]    1    1    1    1    1    1    1    1    1     1
   [3,]    1    1    1    1    1    1    1    1    1     1
   [4,]    1    1    1    1    1    1    1    1    1     1
   [5,]    1    1    1    1    1    1    1    1    1     1

[[2]]
       [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
 [1,]    2    2    2    2    2    2    2    2    2     2
 [2,]    2    2    2    2    2    2    2    2    2     2
 [3,]    2    2    2    2    2    2    2    2    2     2
 [4,]    2    2    2    2    2    2    2    2    2     2
 [5,]    2    2    2    2    2    2    2    2    2     2

Say I want to take the mean across the corresponding elements of the matrices in the list. One way to do this would be 

Reduce("+",A)/length(A)

I can't seem to feed Reduce() more complex functions and assume there is a better way in general.

joran
  • 169,992
  • 32
  • 429
  • 468
scottyaz
  • 722
  • 7
  • 18
  • two lists?... what do you mean by element?... the whole data frame? – John Aug 16 '11 at 21:38
  • Ok, I edited the question to better reflect what you're asking. Hopefully I captured what you're looking for... – joran Aug 16 '11 at 21:42
  • sorry for the confusing language. I want to take the mean of each number in, for example, position [1,1] of each matrix in the list. – scottyaz Aug 16 '11 at 21:44
  • 1
    any function that operates on two objects at a time will work with `Reduce`. `mean` accepts only one R object and that's the reason it fails. i don't think there is a generic approach, but for a class of functions, it would be possible to define a simple wrapper that makes it operate on 2 objects at a time and can be passed to `Reduce`. it would be useful if you could post what function you have in mind – Ramnath Aug 16 '11 at 23:02
  • In general you can't convert a dataframe into an array unless all the data columns are homegeneous (e.g. all integer, as in your example, or all factor, or all string, or all Date). So you would be reduced to the `do.call()` approach. – smci May 16 '12 at 15:18

2 Answers2

7

In this case, maybe you're better off with your data in an array rather than a list?

#Recreate data
A <- list(a=matrix(1,5,10),b=matrix(2,5,10))

#Convert to array
A1 <- array(do.call(cbind,A),dim = c(5,10,2))

#Better way to convert to array
require(abind)
A1 <- abind(A,along = 3)

#Now we can simply use apply
apply(A1,c(1,2),mean)
joran
  • 169,992
  • 32
  • 429
  • 468
  • that's what @hadley [suggested](https://stat.ethz.ch/pipermail/r-help/2008-December/183616.html) to me a while ago – baptiste Aug 16 '11 at 22:12
  • It doesn't change the structure: `apply(abind, A, along=3), 1:2, mean)` does not affect A. – IRTFM Aug 16 '11 at 23:00
  • 1
    Oops, I think that's a minor typo by @DWin: `apply(abind(A,along=3),c(1,2),mean)`. – joran Aug 16 '11 at 23:09
2

Maybe do.call?

do.call(`+`, A)/length(A)

Or if you really don't want to abind it into a larger matrix,

array(sapply(seq_along(A[[1]]), function(i) mean(sapply(A,`[`,i))), 
      dim=dim(A[[1]]))
Aaron left Stack Overflow
  • 36,704
  • 7
  • 77
  • 142
  • that is similar to Reduce() and I run into problems when I supply a more complex function to it (like 'mean') – scottyaz Aug 16 '11 at 21:39
  • Actually `mean` isn't complex enough; this will work on any function that takes multiple arguments and handles them in a vectorized way. Given the further information, though, joran's method of putting it in an array is what you want, though I've added a much clumsier `sapply` solution, which is basically just binding each spot together in turn. – Aaron left Stack Overflow Aug 17 '11 at 03:04