Raul,
I have an alternative method for doing this which is based on a different
approach from im2col and mostly I just want to prove to myself the 2
methods provide identical results. I have done that so, there is not much
reason to pursue this issue now. The different approach is detailed in sec
4.1 at https://arxiv.org/pdf/1603.07285.pdf . (I am not suggesting you look
at this other link,btw.) I do not quite understand your '..."neglected"
dimensions ...', btw.
Still, I thought my first post showed what I need by showing inputs and the
results I want.
I have reproduced those initial results just below to give you what I think
is the solved example. I am trying to provide im2col with a non-standard
(flattened matrice(s), singular as just a list initially, or plural as an
array of one or more matrices) input.
*******reproduced from initial post (below)*********
forcearray =: (($,)~_2{.1,$)
im2col =: (2 2$1 1 2 2) (|:"2@:(,;._3)"2) _4 ]\"1 ]
$,"1&>/"1<"2 im2col forcearray i.12
1 4 6
$,"1&>/"1<"2 im2col forcearray i.2 12
2 4 6
$,"1&>/"1<"2 im2col forcearray i.3 12 NB. etc
3 4 6
,"1&>/"1<"2 im2col forcearray i.2 12
0 1 2 4 5 6
1 2 3 5 6 7
4 5 6 8 9 10
5 6 7 9 10 11
12 13 14 16 17 18
13 14 15 17 18 19
16 17 18 20 21 22
17 18 19 21 22 23
,"1&>/"1<"2 im2col forcearray i. 12
0 1 2 4 5 6
1 2 3 5 6 7
4 5 6 8 9 10
5 6 7 9 10 11
*******reproduced from initial post (above)*********
Thank you,
On Sat, Apr 13, 2019 at 8:56 AM Raul Miller <[email protected]> wrote:
> An you describe more specifically the transform you need with multiple
> matrices?
>
> And/or, can you work through an example where you’re using a 2 3 4 array?
>
> Ignore the code side of it, maybe, initially, if that’s getting in the way?
>
> (I am not quite seeing how im2col fits into this context. But it might be
> that you’ll want to use ,/;._3 with “neglected” dimensions moved so they’re
> left alone at that step?)
>
> Thanks,
>
> —
> Raul
>
> On Saturday, April 13, 2019, Brian Schott <[email protected]> wrote:
>
> > Raul,
> >
> > I like your ideas. You actually got a long way, despite your denial.
> > Please don't spend much time on my question, because I do have working
> > solution.
> > I don't know how to comment on what you are saying about the possible
> > transpose before the ,;._3 being possibly necessary. But such transposes
> > might simplify some of the other issues I am having with my inputs
> matching
> > the later gyrations I describe below.
> >
> > This simplified version of im2col based on your ideas works for me if
> there
> > is just one **matrix** argument. But, see further below, because my
> > argument is not just one matrix.
> >
> > im2col =: (1 1,:2 2)|:@(,/@(,;._3))]
> > im2col i. 3 4
> > 0 1 2 4 5 6
> > 1 2 3 5 6 7
> > 4 5 6 8 9 10
> > 5 6 7 9 10 11
> >
> >
> > But, I think your forcematrix, or my forcearray, is needed because I am
> > supplying one or more *flattened* matrices as lists. So my input to
> im2col
> > is more like B =. i. 2 12 where each item of B is a flattened 3x4 matrix.
> > My previous machinations with boxing and unboxing was largely due to
> > restructuring each item of B to a matrix before giving it to im2col, and
> I
> > still have no simple way to do that and my approach requires ugly
> > processing even after the "sliding" of u;._3 .
> >
> >
> > I can't tell how to gracefully convert arguments like my B into arrays of
> > shapes like 2 3 4 before submitting them to an im2col with revised
> external
> > and internal ranks. Instead my previously posted im2col accepts arguments
> > like B and after the u,._3 reworks the result into the needed form.
> >
> > I think I prefer preprocessing the B into multiple matrices (an array)
> (as
> > suggested in the first sentence of the previous paragraph), but as I
> said,
> > I can't see how.
> >
> > Thanks, again,
> >
> >
>
--
(B=)
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
