Comment on median(x) and NaN

[Ben Abbott]

On Jan 17, 2008, at 12:15 AM, Michael D Godfrey wrote:

> It appears that the Octave sort behavior was done
> to match Matlab.  Thus for
> x= [1,2,3,4,NaN,NaN,NaN,NaN,5,6,7,8,9];
> Octave and Matlab sort(x) return
> 1     2     3     4     5     6     7     8     9   NaN   NaN    
> NaN   NaN
> And, for sort(x,'descend'), both return
>  NaN   NaN   NaN   NaN     9     8     7     6     5     4     3
> 2     1
>
> But, for the above array, or any vector containing NaN,
> Matlab median() returns NaN.  Octave median() returns 7 for
> both median(sort(x)) and median(sort(x,'descend')).  Also,
> mean(x) returns NaN in both Octave and Matlab.
>
> There has been some recent discussion about median(), so this
> observation should be considered in making any changes.
>
> It seems to me that the sort behavior is not consistent with
> the intent of the IEEE spec. However, this may be another case where
> "Matlab compatibility (right or wrong)" may be the better choice,
> at least for now.
>
> The Matlab that I used for this is 7.5.0 (R2007b).
>
> Michael

Good point.

I'm still working on the quantile algorithm. I'll add this *feature*  
to the list.

Ben