[R] else if statement error

[Uwe Ligges]

David Winsemius wrote:
> 
> On Oct 5, 2009, at 5:38 AM, Martin Maechler wrote:
> 
>>>>>>> "DW" == David Winsemius <dwinsemius at comcast.net>
>>>>>>>    on Sat, 3 Oct 2009 12:56:51 -0400 writes:
>>
>>    DW> On Oct 3, 2009, at 11:54 AM, Chen Gu wrote:
>>
>>>> Hello,
>>>>
>>>> I am doing a simple if else statement in R. But it always comes out
>>>> error
>>>> such as 'unexpected error'
>>>> There are two variables. ini and b. when ini=1, a=3; when ini>1 and
>>    b> 2,
>>>> a=5; all other situations, a=6. I don't know where it is wrong.
>>>> Here is my code
>>>>
>>>> ini=3
>>>> b=4
>>    DW> Your basic problem is that you are confusing if and else which are
>>    DW> program control functions with ifelse which is designed for 
>> assignment
>>    DW> purposes;
>>
>> David, not quite:  in R, "everything"(*) is a function,
>> and in the example we have here
>> (and innumerous similar examples)   ifelse(.) is not efficient
>> to use:
>>>> ini=3
>>>> b=4
>>>> a <-   ifelse( ini==1, 3, ifelse( ini>1 & b>2 , 5, 6))
>>>>
>>>> a
>>> [1] 5
>>
>> More efficient -- and also nicer in my eyes ---
>> is
>>     a <-  if(ini == 1) 3 else if(ini > 1 && b > 2) 5  else  6
> 
> No argument that is a bit more readable than my version above. I guessed 
> that:
> 
> a <- 6 - 3*(ini == 1) - (ini > 1 && b > 2)
> 
>  ... might be even more efficient, ... but 10^6 instances took almost 
> twice as long as the if(){}else{} construction. The if else construction 
> took 2e-06 seconds, the ifelse(,,) took 4.2e-05 seconds and the Boolean 
> math version 3.45e-06 seconds.
> 
> 
>>
>> As I say on many occasions:
>>   ifelse() is useful and nice, but it is used in many more
>>   places than it should.... {BTW, not only because of examples like 
>> the above}.
> 
> On the r-help list I see many more examples where (I thought) ifelse 
> should have been used. I would be interested in seeing the other 
> examples that you have in mind where it should not be used. On the 
> principle that there are no new questions on r-help anymore, perhaps a 
> link to a discussion from the archives could be .... more efficient? (I 
> did try searching and have found some interesting material, but not on 
> this particular point. A search for:
> 
> "if else" ifelse efficien*
> 
> came up empty when restricted to R-help.)
> 


The first rule is easy: As long as you are using scalar valued (i.e. 
length 1 vectors in R) "cond", you should prefer
  if(cond) cons.expr  else  alt.expr
rather than
   ifelse(cond, yes, no)
because the latter one evaluates both "yes" and "no" while the former 
one evaluates exactly one of both expressions.

Moreover, if "yes" and "no" are not really trivial, the penalty of 
evaluating both of them should be considered.

Best,
Uwe Ligges




>>
>> Martin Maechler, ETH Zurich
>>
>> ----
>> (*) "almost"
>>
>> -- 
> 
> David Winsemius, MD
> Heritage Laboratories
> West Hartford, CT
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide 
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.