[R] Do NOT use ifelse() if you can use if(.) else . [was "Checking .."]

[Ben Tupper]

On Dec 23, 2014, at 4:05 AM, Martin Maechler <maechler at stat.math.ethz.ch> wrote:

>>>>>> Steven Yen <syen04 at gmail.com>
>>>>>>    on Sun, 14 Dec 2014 09:30:56 -0500 writes:
>>>>>> Steven Yen <syen04 at gmail.com>
>>>>>>    on Sun, 14 Dec 2014 09:30:56 -0500 writes:
> 
>> Thanks. This worked!! :)
>> Fisher <- ifelse(!("Fisher" %in% names(obj$spec)), FALSE, obj$spec$Fisher)
> 
> "worked",  yes.
> 
> But please --- for the mailing list archives ---
> do use better code.
> 
> Unfortunately, ifelse() is used and has been advertized much too
> often in cases where it is very sub-optimal efficiency wise. 
> ifelse(Cond, A, B)  should only be used  when the condition
> 'Cond', a logical vector can be (and typically is) of length > 1
> {and even then, it maybe a nice short cut, but often still suboptimal
> efficiency wise ... but let's not get there}
> 
> In cases like this one when the condition 'Cond' is a
> simple TRUE or FALSE (i.e. of length 1), using
> 
> 	if(Cond) A else B
> 
> instead of
> 
> 	ifelse(Cond, A, B)
> 
> is
> 
>  1. much more R - like  [[ "everything you do is a function call" ]]
>     hence much more elegant
> 
>  2. considerably more efficient.
> 
>  3. :-) less typing: uses two "," less ;-)
> 
>> require(microbenchmark)
>  Loading required package: microbenchmark
>> x <- setNames(,LETTERS)
>> y <- setNames(letters, runif(letters))
>> z <- pi
>> microbenchmark(r1 <- ifelse(z > 3, x, y), r2 <- if(z > 3) x else y, times=1000)
>  Unit: nanoseconds
> 			expr  min     lq     mean median     uq   max neval cld
>   r1 <- ifelse(z > 3, x, y) 4466 4971.5 5498.928   5244 5673.5 31705  1000   b
>   r2 <- if (z > 3) x else y  171  212.0  265.241    264  291.0  3130  1000  a 
>> 
> 
> i.e., roughly a factor of 20 times more efficient
> 

Wow! This is great, and I had no idea I was misusing ifelse. 

Thanks!
Ben



> 
> Martin Maechler,
> ETH Zurich and R Core Team.
> 
> 
>> At 08:43 AM 12/14/2014, Ben Tupper wrote:
>>> Hi,
>>> 
>>> Does this work for you?  It simply tests if the name Fisher is found 
>>> among the names of the elements of spec.
>>> 
>>> obj = list(spec = list)
>>> Fisher <- ifelse(!("Fisher" %in% names(obj$spec)), FALSE, obj$spec$Fisher)
>>> 
>>> Cheers,
>>> Ben
>>> 
>>> On Dec 14, 2014, at 8:07 AM, Steven Yen <syen04 at gmail.com> wrote:
>>> 
>>>> My obj does not always come with a logical variable defined. So I do
>>>> 
>>>> my.foo <- function(obj,df,digits=5){
>>>> if (!is.na("obj$spec$Fisher")) Fisher<-obj$spec$Fisher
>>>> ...
>>>> }
>>>> 
>>>> This works when "Fisher" is defined in/passed from obj. When it 
>>> is not, I get error:
>>>> 
>>>> Error in (!is.na("obj$spec$Fisher")) & Fisher :
>>>> operations are possible only for numeric, logical or complex types
>>>> 
>>>> I tried exist(Fisher), missing(Fisher)... to no vail. Any idea? Thanks.

Ben Tupper
Bigelow Laboratory for Ocean Sciences
60 Bigelow Drive, P.O. Box 380
East Boothbay, Maine 04544
http://www.bigelow.org