R devel - Oct 2009 - inconsistency in return value of peaks() {splus2R} (PR#13988)

Full_Name: Benny van der Vijgh
Version: 2.7.2
OS: Windows Vista
Submission from: (NULL) (194.171.252.108)


The return value of peaks() in package splus2R is not consistent. 
This is because of the call to max.col() without additional parameters which
peaks() makes.
max.col() has a parameter 'ties.method' which specifies how ties are
handled,
with "random" by default. 
This means that when peaks() is called on a vector with two (or more)
neighbouring elements in a tie for the local maximum, these end up in the same
row in the embedded matrix after which max.col() at random chooses which will be
considered the maximum, thereby yielding inconsistent results, for example:
> a <- c(3,5,5,3)
> peaks(a)
[1] FALSE FALSE FALSE FALSE
> peaks(a)
[1] FALSE  TRUE FALSE FALSE

this can be easily fixed by setting the parameter 'ties.method' for
max.col() to
"first" or "last", which is a matter of application and
taste I think, perhaps
it can even be considered to make this parameter also a parameter of peaks() and
pass it to max.col() in the execution.

vijgh at phil.uu.nl wrote:
> Full_Name: Benny van der Vijgh
> Version: 2.7.2
> OS: Windows Vista
> Submission from: (NULL) (194.171.252.108)
>
>
> The return value of peaks() in package splus2R is not consistent. 
>   
This is not an R bug.  Please contact the author of that package.

Duncan Murdoch
> This is because of the call to max.col() without additional parameters
which
> peaks() makes.
> max.col() has a parameter 'ties.method' which specifies how ties
are handled,
> with "random" by default. 
> This means that when peaks() is called on a vector with two (or more)
> neighbouring elements in a tie for the local maximum, these end up in the
same
> row in the embedded matrix after which max.col() at random chooses which
will be
> considered the maximum, thereby yielding inconsistent results, for example:
>
>   
>> a <- c(3,5,5,3)
>> peaks(a)
>>     
> [1] FALSE FALSE FALSE FALSE
>   
>> peaks(a)
>>     
> [1] FALSE  TRUE FALSE FALSE
>
> this can be easily fixed by setting the parameter 'ties.method' for
max.col() to
> "first" or "last", which is a matter of application and
taste I think, perhaps
> it can even be considered to make this parameter also a parameter of
peaks() and
> pass it to max.col() in the execution.
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>