It seems to me that the matrix classes are using the object functions
in a fairly clean way. If you are doing, well, a reduction over a
matrix<T>, you need a function that takes two T's and returns a T.
Assuming, of course, that matrices of Objects are any more useful then
bicycles for lobsters. If you want me to put cement overshows on the
whole concept of an Object matrix, that's fine. I can't find any
evidence that they are used, but I'm not 100% sure.
I propose as follows: (1) commit what I've got, and then try the
experiment of blowing off the object matrices and see if anything is
left to squeak.
On Sun, Jan 17, 2010 at 8:56 PM, Ted Dunning <ted.dunn...@gmail.com> wrote:
> I looked at the patch here and wondered if all of this code could just as
> well be lost rather than generifying it. In particular, the ObjectFunction
> assumption that argument and result types are the same looks like trouble
> waiting to happen. Deleting the code won't cause any more trouble to a
> prospective user than leaving it with implausible assumptions that we can't
> fathom yet.
>
> On Sun, Jan 17, 2010 at 5:20 PM, Benson Margulies
> <bimargul...@gmail.com>wrote:
>
>> I think that other eyeballs should be applied, however briefly and
>> painfully, to MAHOUT-256.
>>
>> In addition, I propose to REMOVE the ObjectFactory{1,2,3}D classes. No
>> one uses them, and they would require significant API surgery to be
>> genric-cleaned. Since they have protected constructors and no static
>> methods, it seems reasonable to apply the surgeon's motto as
>> previously discussed.
>>
>
>
>
> --
> Ted Dunning, CTO
> DeepDyve
>
