You can avoid the overhead of memory allocation for the ( a,:b ) part by
recognizing that ( c{ ) can access "a" and "b" in-place without building
the joined array. Why this happens for the assignment case and not
generally, I don't know but I suspect Raul is on the right track with the
insight that idiom recognition needs to be simple to avoid overhead for the
non-idiomatic cases.
However, when you're parsing the assignment case, you'd know that you're
going to allocate non-temporary memory - which incurs a potentially large
penalty - and this might be a cue to spend more time on idiom recognition
(which is kind of what Raul said).
On Thu, Dec 22, 2011 at 12:58 PM, David Vaughan <
purpleblue...@googlemail.com> wrote:
> Why should there be any speed up just from assigning the phrase to a noun?
> Surely in both cases the same code has to be executed, one will display the
> result, one will store the result?
>
> ___________________________
>
> David Vaughan
>
> On 22 Dec 2011, at 17:46, Raul Miller <rauldmil...@gmail.com> wrote:
>
> > Actually, I think he was asking why the speedup is not available without
> the =:
> >
> > And I think that has to do with "ease of implementation".
> >
> > Also, if there's no =: then that syntax is almost certainly being used
> > to generate a noun -- it's not being used to build a repeatable
> > operation. You'd have to use different syntax to generate a useful
> > verb.
> >
> > --
> > Raul
> >
> > On Thu, Dec 22, 2011 at 12:24 PM, Devon McCormick <devon...@gmail.com>
> wrote:
> >> In a word: performance.
> >>
> >> a=. 1e6?@$0
> >> b=. <:+:1e6?@$0
> >> c=. 1e6?@$2
> >>
> >> 10 ts 'r=. c}b,:a'
> >> 0.0066557413 8390912
> >> 10 ts 'c}b,:a'
> >> 0.021307987 41945088
> >>
> >>
> >> On Thu, Dec 22, 2011 at 11:00 AM, Johann Hibschman <
> jhibsch...@gmail.com>wrote:
> >>
> >>> Thanks, everyone. I never understood item-amend until now. That's
> >>> exactly what I was looking for. For context, I was reading through the
> >>> APL book and hit the "/a,u,b/" notation, and wondered how to do that
> >>> in J.
> >>>
> >>> One question though: why is the particular form (result=:c}b,:a)
> >>> singled out, rather than the general (c}b,:a)?
> >>>
> >>> On Wed, Dec 21, 2011 at 10:54 AM, Henry Rich <henryhr...@nc.rr.com>
> wrote:
> >>>> Note that
> >>>>
> >>>> c}b,:a
> >>>>
> >>>> does indeed compute b,:a. But if it's used in the precise form
> >>>>
> >>>> result =: c}b,:a
> >>>>
> >>>> the computation is avoided.
> >>>>
> >>>> b =: 1e6$4
> >>>> a =: 1e6$4
> >>>> c=: ? 1e6$ 2
> >>>> ts 'c}b,:a'
> >>>> 0.0401299 2.09725e7
> >>>> ts 'q =: c}b,:a'
> >>>> 0.0112949 4.19558e6
> >>>>
> >>>> There are other special cases; see the Special Code page.
> >>>>
> >>>> Henry Rich
> >>>>
> >>>> On 12/21/2011 11:43 AM, Marshall Lochbaum wrote:
> >>>>> A more obvious way is to use (c { b,.a) . The key is to realize that
> >>> saying
> >>>>> (if c then a else b) is just selection: if c is 0, we pick b, and if
> c
> >>> is
> >>>>> 1, we pick a. So we rephrase as (c { b,a), and J's implicit rank
> allows
> >>> us
> >>>>> to do (c { b,.a) for vector c, a, and b.
> >>>>>
> >>>>> That said, (c} b,:a) is faster, and uses less memory by not actually
> >>>>> computing b,:a . So it's a better choice.
> >>>>>
> >>>>> Marshall
> >>>>>
> >>>>> On Wed, Dec 21, 2011 at 10:26 AM, Brian Schott<
> schott.br...@gmail.com
> >>>> wrote:
> >>>>>
> >>>>>> One way is the following.
> >>>>>>
> >>>>>> c}b,:a
> >>>>>> 10 2 30
> >>>>>>
> >>>>>>
> >>>>>> --
> >>>>>> (B=)
> >>>>>>
> ----------------------------------------------------------------------
> >>>>>> 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
> >>>
> >>
> >>
> >>
> >> --
> >> Devon McCormick, CFA
> >> ^me^ at acm.
> >> org is my
> >> preferred e-mail
> >> ----------------------------------------------------------------------
> >> 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
>
--
Devon McCormick, CFA
^me^ at acm.
org is my
preferred e-mail
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
