Skip to content

0/1 vector for indexing leads to funny behaviour (PR#8389)

8 messages · rasche@molgen.mpg.de, Peter Dalgaard, Tony Plate +3 more

rasche@molgen.mpg.de
# 
Full_Name: Axel Rasche
Version: 2.2.0
OS: Linux
Submission from: (NULL) (141.14.21.81)


Dear Debuggers,

This is not a serious problem. Are 0/1 vectors intended to be used as index
vectors? If yes, there is a bug. If not, it leads just to some funny behaviour
rather than an error message.

In the appendix is some simple code to reproduce the problem. A logical vector
as.logic(a) helps by indexing the vector b. The 0/1 vector a just returns the
first value "a". But as many times as there is a 1 in a.

Best regards,
Axel


Appendix:

b = c("a","b","c","d")
a = c(0,1,1,0)
b[as.logical(a)]
b[a]
a = c(1,0,1,0)
b[as.logical(a)]
b[a]
a = c(0,1,1,1)
b[as.logical(a)]
b[a]
Peter Dalgaard
# 
rasche at molgen.mpg.de writes:
Yes, that is completely as intended. Zeros in a numerical index vector
produce nothing and ones produce the first element. The documentation
could arguably be better on this point though.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Tony Plate
    

    

      
      #
    

  


  

      
Yes, 0/1 (numeric) are intended to be used as index vectors -- and they 
have the semantics of numeric indices, which is that 0 elements in the 
index are omitted from the result.  This can be a very useful mode of 
operation in many situations.

I was going to write "This is described in both the introduction to R, 
and in the documentation for '['", except that I checked before I wrote 
and was surprised to be unable to any discussion of zeros in indexing in 
any of the first three places I looked:

(1) help page for '[' (There is discussion of zero indices here, but 
only in the context of using matrices to index matrices, not in the 
context of ordinary vector indices).

(2) Section 2.7 "Index vectors: selecting and modifying subsets of a 
data set" in "An Introduction to R", which does say this about numeric 
indices:
     2. A vector of positive integral quantities. In
        this case the values in the index vector must
        lie in the set {1, 2, . . . , length(x)}
(This seems to commit the sin of not telling the whole truth.)

(3) Section 5.5 "Array Indexing.  Subsections of an array" (In "An 
Introduction to R")

Question for others: did I miss something obvious, or is this a 
documentation deficiency that zeros in indices are not discussed in 3 of 
some obvious first places to look?

If indeed this is a documentation deficiency, I'm happy to contribute 
documentation patch, but I await other opinions before spending any time 
on that.

-- Tony Plate

        
rasche at molgen.mpg.de wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Gabor Grothendieck
    

    

      
      #
    

  


  

      
The other place its discussed is in 3.4.1 of the R Language Definition:

http://finzi.psych.upenn.edu/R/doc/manual/R-lang.html#Indexing-by-vectors

        
On 12/13/05, Tony Plate <tplate at acm.org> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
?"[" says

See Also:

      'list', 'array', 'matrix'.

      '[.data.frame' and '[.factor' for the behaviour when applied to
      data.frame and factors.

      'Syntax' for operator precedence, and the _R Language_ reference
      manual about indexing details.

and the `indexing details' are indeed where it says they are.

This is not an introductory topic, and it makes sense to have the details 
in only one place and refer to it.  That help page is already over-loaded.

        
On Tue, 13 Dec 2005, Tony Plate wrote:

        

            

      
      
      
    

        
No. Zero is not a positive integer.

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Tony Plate
    

    

      
      #
    

  


  

      
I appreciate the explanation that some details should not appear in the 
help pages or the Introduction to R manual.

However, I am puzzled by this part of Prof Ripley's response:

TP> [...] "An Introduction to R" [...] says this about
TP> numeric indices:
TP>     2. A vector of positive integral quantities. In
TP>        this case the values in the index vector must
TP>        lie in the set {1, 2, . . . , length(x)}
TP> (This seems to commit the sin of not telling the whole truth.)

BDR> No. Zero is not a positive integer.

That's what I was trying to say: the whole truth is that numeric index 
vectors that contain positive integral quantities can also contain 
zeros.  Upon rereading this passage yet again, I think it is more 
misleading than merely incomplete: the phrasings "positive integral 
quantities", and "*must* lie in the set ..." rule out the possibility of 
the vector containing zeros.

In this Section 2.7 in "An Introduction to R", the four types of index 
vectors are introduced with "Such index vectors can be any of four 
distinct types:". There is not even a hint that other types of index 
vectors can be used (e.g., positive integral quantities and zeros).  Is 
this really correct and helpful?  (The only way that I can see that this 
section can be interpreted as correct is to claim that that the phrasing 
"can be any of four distinct types" permits the existence of other types 
that are neither described nor hinted at.  However, this interpretation 
feels more clever than helpful.)

Tony Plate

        
Prof Brian Ripley wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Peter Dalgaard
    

    

      
      #
    

  


  

      
Tony Plate <tplate at acm.org> writes:

            

      
      
      
    

        
I wondered too. I suppose one interpretation is, No, it's not just not
telling the whole truth, it's outright false! Alternatively, it could
be that 2. is OK as written, but there needs to be entries for the
nonnegative/nonpositive cases. Or - perish the thought - that Brian
made a blunder...

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Thomas Lumley
    

    

      
      #
    

  


  

      
On Wed, 14 Dec 2005, Tony Plate wrote:

            

      
      
      
    

        
"Someone told me that you can't run without bouncing the ball in 
basketball. I got a basketball and tried it and it worked fine. He must be 
wrong"  -- a comp.lang.c standard

It doesn't rule out the the possibility of the vector containing zeros, it 
tells you that you should not put zeros in the vector.

 	-thomas