Re: package nan warnings

[Alois Schloegl]

On 2012-08-02 21:53, Jordi Gutiérrez Hermoso wrote:
> On 2 August 2012 15:42, Yin, Yue-Jun<address@hidden>  wrote:
> > Alois,
> > Thank you. I got your point about the agreement issue.
> > What is the consequence of shadowing? The performance is enhanced or the other 
> > way?
>
> The NaN package overwrites core functions, because Alois thinks
> they're buggy. I disagree. The reason to overwrite them is that Alois
> thinks NaNs should never be propagated in a calculation. The point of
> the NaN package is to do this, to never propagate NaNs, but just omit
> them from a calculation. So for example, in core Octave sum([1 NaN 2])
> is NaN, but in Alois's package it's 3.
>
> - Jordi G. H.

Jordi,

you still do not get it.

1) I've never stated that this core functions are "buggy" - Never. 
However, I'm saying that there is a better way to deal with NaN's that 
what these "core functions" do.
2) The idea of the NaN-toolbox is not that "NaNs should never be 
propagated in a calculation" but that NaN's should be ignored in 
*statistical" analysis (emphasis on "statistical").
3) after installing the NaN-toolbox,  sum([1 NaN 2]) will still result 
in NaN. But with the NaN-toolbox you have an additional function 
sumskipnan([1,NaN,2]) which gives 3.

Conclusion, you are wrong in all of your three accounts.


Alois