Re: nchoosek

[Daniel J Sebald]

On 08/31/2014 08:35 PM, Rik wrote:
> On 08/31/2014 09:00 AM, address@hidden wrote:
> > Le 31/08/2014 05:43, Rik a écrit :
[snip]
> However, when the first input is a vector, the input is treated as a set.
> nchoosek returns all possible subsets of the original set which have the
> desired property of being a unique combination of the original elements in
> groups of k.
>
> nchoosek ([4 5], 8)
> is not
> [ nchoosek(4, 8), nchoosek(5,8) ]
>
> so I don't think it should return [0, 0].

Plus, if it did return [0 0] then there would be a non-unique output because

octave-cli:1> nchoosek([0 0], 2)
ans =

   0   0

Interpreting the behavior based upon vector/scalar has always been a 
dodgy method because of the inherent restrictions it imposes.


> Right now we return the empty set, which is true because that is the only
> subset that matches.  I do find it odd that we give an error in the first
> case and not in the second.  If I had to vote I would change the first case
> to return 0 since that is close to the second case of returning the empty
> set [].

Julien, you mentioned commonly using nchoosek-like functions for k 
outside the range [0 N].  Can you give an example?  And what if k is 
negative?  Error or 0?

Dan