for,
,/^:(2 -~ #@$)^:(3 <: #@$)
u^:(3 <: #@$)
is that 0/1 form, and u does not get executed if 0?
Correct.
,/^:(2 -~ #@$)
is ,/^:n (where n could be greater than 1)
on timex tests with input (i. 1 2 3 7 4 5 2 3 3 4 5) all of the options are
super close.
They are going to be close because they are creating self-virtual blocks
which is very fast, with no data movement.
Bob's version (^:_) is probably the fastest for 3 shape counts.
Not so fast.
NB. Create an inplaceable array
7!:2 '2 2 2 100 100 ?@$ 0'
1049792
NB. Now let's combine the first 4 axes:
7!:2 ',/^:(2< #@$)^:(_) 2 2 2 100 100 ?@$ 0'
1051712
7!:2 '4 (((*/@{. , }.) $) ($,) ]) 2 2 2 100 100 ?@$ 0'
1051840
NB. Which takes less space?
NB. Are you sure?
7!:2 'a =. ,/^:(2< #@$)^:(_) 2 2 2 100 100 ?@$ 0'
2100160
7!:2 'a =. 4 (((*/@{. , }.) $) ($,) ]) 2 2 2 100 100 ?@$ 0'
1052224
Surprised?
7!:2 'a =. ,/^:(2 -~ #@$) 2 2 2 100 100 ?@$ 0'
1051648
u^:_ cannot run inplace, because each input to u must be saved for
comparison with the result. Thus, the operation produces virtual
blocks, but not self-virtual blocks, which means that when the result is
assigned to a name (or boxed) the value must be realized.
u^:n for finite n CAN run inplace, and when u is ,/ it produces a
self-virtual block, which can be assigned without being realized.
[a self-virtual block is a virtual block that can be written on top of
its backing block, because the backing block is inplaceable and the new
shape is no longer than the old. (,/ y) creates a self-virtual block,
as does (1000 {. i. 1024) .]
Henry Rich
On Sunday, May 24, 2020, 12:59:07 p.m. EDT, Henry Rich <[email protected]>
wrote:
Counting comparisons is SO not the right way to look at J. The
interpreter is doing much computation behind the scenes. Here, in the
original interpreter, executing u^:v causes v to be executed to create a
noun n; then u^:n is reinterpreted as a conjunction to produce a verb;
then that verb gets executed. I modified this to avoid the
reinterpretation if v produces atomic 0 or 1. Meanwhile, what really
makes a difference is whether the result will be a virtual block or, if
possible, an in-place (self-virtual) block. This difference will show
up if you assign the result.
It turns out that, because of some recent rewriting, ,/^:(2 -~ #@$)^:(3
<: #@$) does execute ,/ inplace, and ,/ does produce a self-virtual
block, twice, and so this idiom takes not much more space than
2 (((*/@{. , }.) $) ($,) ]) y
which is how I would write it.
Counting cycles is quite difficult nowadays even when you are looking at
the assembler code; impossible from the J code unless you know the
internals.
Henry Rich
On 5/24/2020 10:45 AM, 'Pascal Jasmin' via Programming wrote:
to only "reduce" if rank 3 or higher, as Bob's version did
,/^:(2 -~ #@$)^:(3 <: #@$) i. 4 5
the advantage is only one comparison is needed if shape count is under 3, and
exactly 2 comparisons if more than 2.
,/^:(2 -~ #@$) i. 1 2 3 4 5
On Saturday, May 23, 2020, 11:36:20 p.m. EDT, 'robert therriault' via Programming
<[email protected]> wrote:
Hi Bill,
You could use this
,/^:(2< #@$)^:_
$ ,/^:(2< #@$)^:_ i. 5
5
$ ,/^:(2< #@$)^:_ i. 4 5
4 5
$ ,/^:(2< #@$)^:_ i. 3 4 5
12 5
$ ,/^:(2< #@$)^:_ i. 2 3 4 5
24 5
$ ,/^:(2< #@$)^:_ i. 7 2 3 4 5
168 5
although Roger or Pascal may have something more elegant.
Cheers, bob
On May 23, 2020, at 20:09, bill lam <[email protected]> wrote:
Wow so simple!
Thanks Roger, Pascal.
Just curious, how to do rank-n to rank-2
where n>3
On Sun, May 24, 2020, 11:04 AM 'Pascal Jasmin' via Programming <
[email protected]> wrote:
,/ i. 3 4 5
On Saturday, May 23, 2020, 10:56:02 p.m. EDT, bill lam <
[email protected]> wrote:
I had a problem similar to that of the recent shape question,
a=. i. 3 4 5
to reshape to rank-2 array with trailing shape unchanged that is shape 12
5,
This works but difficult to remember
(((*/@}: , {: )@: $) $ , ) i. 3 4 5
0 1 2 3 4
5 6 7 8 9
10 11 12 13 14
15 16 17 18 19
20 21 22 23 24
25 26 27 28 29
30 31 32 33 34
35 36 37 38 39
40 41 42 43 44
45 46 47 48 49
50 51 52 53 54
55 56 57 58 59
this can also work and easier to remember but Henry Rich said recently
that ]\ doesn't support virtual block yet.
_5 ]\ ,i. 3 4 5
0 1 2 3 4
5 6 7 8 9
10 11 12 13 14
15 16 17 18 19
20 21 22 23 24
25 26 27 28 29
30 31 32 33 34
35 36 37 38 39
40 41 42 43 44
45 46 47 48 49
50 51 52 53 54
55 56 57 58 59
Any suggestions of other ways to do it?
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
--
This email has been checked for viruses by AVG.
https://www.avg.com
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
