> Van: [email protected] [mailto:programming-
> [email protected]] Namens Raul Miller
(...)
> 
> On Wed, Mar 3, 2010 at 2:20 PM, R.E. Boss <[email protected]> wrote:
> > No.
> >
> >   <oblq i.2 3 4
> > |assertion failure
> > |   (=<.)rnk2=:-:#$y
> 
> I explicitly left odd ranks unhandled because I had not yet
> decided how I should deal with them.
> 
> That said, note that ,: will convert an odd ranked array to
> an even ranked array.
> 
> So you might replace that first line with
>    rnk2=: <.-:#$y
> 
> That said, note that my model of "oblique" will
> be incompatible with the existing /. for higher ranked
> arrays.  Given how it works, it might be desirable to
> ignore trailing dimensions in y corresponding to the
> monadic rank of u.
> 
> But I was just modelling the underlying mechanism, and
> not providing a complete implementation.
> 
> > obl=: 1 : '(+/"1)@(#: [:i. */)@$ u/. ,'
> 
> Consider this case:
> 
>   (i.2 3 5) +/oblq@(*/) i.1 2 3
> 


My proposal for a generalization of the monad /. (oblique) was selecting the
hyper planes with atoms which coordinates sum up to a constant number.
Your proposal is the hyper planes with atoms which odd coordinates plus its
even coordinates add up to a constant set of numbers. This one is highly
related to convolutions.
In the 2-dimensional case, these proposals are both equal to /. (oblique) .

   <oblq i.2 3 3 2 
+----+----------+-----------+-----+
|0   |1 6       |7 12       |13   |
+----+----------+-----------+-----+
|2 18|3 8 19 24 |9 14 25 30 |15 31|
+----+----------+-----------+-----+
|4 20|5 10 21 26|11 16 27 32|17 33|
+----+----------+-----------+-----+
|22  |23 28     |29 34      |35   |
+----+----------+-----------+-----+

   <obl i.2 3 3 2 
+-+--------+-------------------+---------------------------+----------------
-------+-----------+--+
|0|1 2 6 18|3 4 7 8 12 19 20 24|5 9 10 13 14 21 22 25 26 30|11 15 16 23 27
28 31 32|17 29 33 34|35|
+-+--------+-------------------+---------------------------+----------------
-------+-----------+--+


R.E. Boss

----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to