[R] extract positions from matrix

[David Winsemius]

On Nov 19, 2011, at 1:15 PM, Bert Gunter wrote:

> Folks:
>
> David: I believe your approach needs to be modified to accommodate
> hundreds of matrices per the OP's specification.

I suggested that my approach was most applicable if the questioner had  
his matrices already in a list. (He didn't say, so neither of us knows  
yet what his/her use case is.)

>
> However, I would say that this is exactly the situation in which
> arrays should be used. Not only does it appear simpler, but it also
> could be faster since once the array is created, (vectorized) indexing
> is used. The key is to take advantage of column major ordering of
> arrays in R as follows:
>
> 1. The OP's example specification is wrong -- he wants positions 1 and
> 5 of the first COLUMN of each matrix (which both solutions gave, but
> just didn't mention).
>
> 2. The code below creates the array assuming that the global workspace
> has only the matrices in it. Subscript ls() appropriately if this is
> not the case.
>
>> for(i in 1:3) assign(LETTERS[i], 15*(i-1)+seq_len(15)) ## toy data
>> rm(i)  ## want only the matrices in the global workspace
>
>> ar <-  array( sapply(ls(), get), dim=c(5,3,3))  ## the dim argument  
>> could also be specified programatically
>

I wonder if you are doing ".jpg" a favor by putting ls() at the center  
of your strategy unless you offer the tools to isolate these objects  
from the rest of the global environment. Who works with no other  
objects in the workspace?

Perhaps this would add the extra level of encapsulation needed for the  
sapply(ls() method of access the created matrices in a probably less  
disruptive fashion:

e <- new.env()
with(e, { for(i in 1:3) assign( LETTERS[i], 15*(i-1)+seq_len(15) )
rm(i)
ar <-  array( sapply(ls(), get), dim=c(5,3,3)) })

# ----------------
e$ar[c(1,5),1, ]

      [,1] [,2] [,3]
[1,]    1   16   31
[2,]    5   20   35

-- 
David.



>> ar[c(1,5),1,]
>
>     [,1] [,2] [,3]
> [1,]    1   16   31
> [2,]    5   20   35
>
>
> Cheers,
> Bert
>
> On Sat, Nov 19, 2011 at 9:19 AM, David Winsemius <dwinsemius at comcast.net 
> > wrote:
>>
>> On Nov 19, 2011, at 9:32 AM, R. Michael Weylandt wrote:
>>
>>> Here's one approach:
>>>
>>> A=matrix(1:15,5)
>>> B=matrix(15:29,5)
>>> C=matrix(30:44,5)
>>>
>>> do.call(cbind, lapply(c("A","B","C"),function(x) get(x)[c(1,5),1]))
>>
>> Also:
>>
>> sapply( list(A,B,C), function(x) do.call("[", list(x, c(1,5)))  )
>>
>> Notice that this actually was extracting using what might be called  
>> the
>> "vector positions". If you wanted to use the i,j version of "["  
>> then you
>> would need an extra column (this example pulling the second columns  
>> in the
>> rows selected:
>>
>>> sapply(list(A,B,C),function(x) do.call("[", list(x, c(1,5), 2)) )
>>     [,1] [,2] [,3]
>> [1,]    6   20   35
>> [2,]   10   24   39
>>
>>
>> Comments on the differences: Michaels version used cbind to get the  
>> ruslt in
>> a matrix, whereas mine used sapply. His used the get function to  
>> extract the
>> objects from a character vector, whereas mine never constructed a  
>> character
>> vector. Which one you deploy will depend on your data setup. If you  
>> already
>> have these in a list, mine might be easier, but if they constitute  
>> an easily
>> constructed set of names you might use LETTERS[] numbers and  
>> paste() to
>> build your list of object names.
>>
>> --
>> david.
>>
>>
>>
>>>
>>> Michael
>>>
>>> On Thu, Nov 17, 2011 at 9:44 PM, .Jpg <jporobicg at gmail.com> wrote:
>>>>
>>>> Hi everyone, I tried to solve this problem but I could not find the
>>>> solution. I have about 105 matrices of equal size in the memory  
>>>> of**R, I
>>>> need to do is extract from these matrices some known positions and
>>>> create a new matrix with these columns. Show you an example with  
>>>> only
>>>> three matrices (but in my case I have hundreds of them).
>>>>
>>>> A=matrix(1:15,5)
>>>>
>>>> [,1] [,2] [,3]
>>>> [1,] 1 6 11
>>>> [2,] 2 7 12
>>>> [3,] 3 8 13
>>>> [4,] 4 9 14
>>>> [5,] 5 10 15
>>>>
>>>> B=matrix(15:29,5)
>>>> [,1] [,2] [,3]
>>>> [1,] 15 20 25
>>>> [2,] 16 21 26
>>>> [3,] 17 22 27
>>>> [4,] 18 23 28
>>>> [5,] 19 24 29
>>>>
>>>> C=matrix(30:44,5)
>>>> [,1] [,2] [,3]
>>>> [1,] 30 35 40
>>>> [2,] 31 36 41
>>>> [3,] 32 37 42
>>>> [4,] 33 38 43
>>>> [5,] 34 39 44
>>>>
>>>> The positions I wish to extract are 1 and 5 of the first row of  
>>>> each
>>>> matrix (in my case are 25positions) and with this generate a new  
>>>> matrix
>>>> with the form
>>>>
>>>> d=
>>>>
>>>> [,1] [,2] [,3]
>>>> [1,] 1 15 30
>>>> [2,] 5 19 34
>>>>
>>>> The ideais to builda loop toextract thisinformation from
>>>> hundredsmatrices, butI failed todo so.
>>>>
>>>> Any helpwould be greatthank you very muchin advance
>>>>
>>>> regards
>>>> .jpg
>>>>
>>
>
>
>
> -- 
>
> Bert Gunter
> Genentech Nonclinical Biostatistics
>
> Internal Contact Info:
> Phone: 467-7374
> Website:
> http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm

David Winsemius, MD
West Hartford, CT