Hi,
I wonder how in J one should properly solve the following problem.
Basically, one need to apply a dyad of rank 0 0 to equally long
arrays of rank 1, with one array being calculated on the fly and
having a variable length. Is it possible to avoid both boxing and
the necessity to handle filling?
Suppose one has a dyad f1, which takes atoms as both arguments and
returns a list, rank 1, of variable length, for example
f1 =: 4 : 'y + i. x' "0 0
Suppose then one has another dyad, f2, also taking atoms as both
arguments, like this:
f2 =: 4 : 'x , y' "0 0
If one wants to apply f1 to 2 and 3, and then apply f2 to the
result of the previous operation and 3, he can write
f2 & (3) 2 f1 3
If one calls f1 with left array of rank 1 and right atom, he gets
an array of rank 2, with fillers to make all partial results of
the same size, like in
2 3 f1 4
4 5 0
4 5 6
Suppose then he wants to apply f2 to such a result as y, with left
x argument - consecutive items from a rank 1 array. Like this:
f2 & (3 4) 2 3 f1 4
4 3
5 3
0 3
4 4
5 4
6 4
But if he doesn't want to deal with the filler 0, which appeared in
the top line of the f1 result, what he has to do? He can box partial
results and then work with accumulated result, like here -
3 4 (f2~ >) "(0 0) 2 3 < @ f1 4
4 3
5 3
0 0
4 4
5 4
6 4
So, here he has a boxing and unboxing only to avoid the fill.
Is boxing justified here? Or is there another way to do what he needs?
Thank you,
Alexander
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
