I think you need to take more care in doing benchmarks. What I got just
now:
10 timer 'idMate1 9000'
0.110233
10 timer 'idMatt1 9000'
0.125659
10 timer 'idMatt2 9000'
0.106581
10 timer 'idMatt3 9000'
0.120995
On Tue, Nov 20, 2012 at 5:48 PM, bob therriault <bobtherria...@mac.com>wrote:
> Wow, would not have expected that , Devon :)
>
> After testing a substantially reduced challenge on my laptop (Devon, you
> have some serious metal behind your J installation!)
> idMate1=: 3 : '(2$y)$(>:y){.1' NB. Explicit and fast
> (10) 6!:2 'idMate1 &.> 9000 '
> 0.120056
> idMatt1=: (2$])$1{.~>: NB. Tacit and fast
> (10) 6!:2 'idMatt1 &.> 9000 '
> 0.11531
> idMatt2=: ,~ $ >: {. 1: NB. Tacit and slow
> (10) 6!:2 'idMatt2 &.> 9000 '
> 0.510053
>
> I am guessing that the 1: is the culprit and that these verbs may be used
> rarely enough that they don't get the benefit of special code
>
> idMatt3=: ,~ $ 1{.~ >: NB. Same as idMatt2 without use of 1:
> (10) 6!:2 'idMatt3 9000 '
> 0.114926
>
> Thanks for the insight Devon, as I await Roger's clarification if I have
> guessed wrong. :)
>
> Cheers, bob
>
> On 2012-11-20, at 5:12 PM, Devon McCormick wrote:
>
> > Here's what I get for timings for some of these methods:
> >
> > NB. Various ways to create ] x ] identity mat.
> > idMat0=: =@:i.
> > idMat1=: 3 : '(2$y)$(>:y){.1' NB.* faster, scales better
> > idMat2=: ,~ $ >: {. 1:
> >
> > (10) 6!:2 'idMat0 &.> 36000 38000 40000 42000 44000 46000 48000 50000'
> > 21.7756
> > (10) 6!:2 'idMat1 &.> 36000 38000 40000 42000 44000 46000 48000 50000'
> > 9.55495
> > (10) 6!:2 'idMat2 &.> 36000 38000 40000 42000 44000 46000 48000 50000'
> > 23.0792
> >
> > If you need a *:5e4 identity matrix, you're in trouble.
> >
> >
> > On Tue, Nov 20, 2012 at 6:04 PM, bob therriault <bobtherria...@mac.com
> >wrote:
> >
> >> Devon,
> >>
> >> I like that, or another tacit option
> >>
> >> idMat1=: ,~ $ >: {. 1:
> >> idMat1 6
> >> 1 0 0 0 0 0
> >> 0 1 0 0 0 0
> >> 0 0 1 0 0 0
> >> 0 0 0 1 0 0
> >> 0 0 0 0 1 0
> >> 0 0 0 0 0 1
> >>
> >> Cheers, bob
> >>
> >> On 2012-11-20, at 1:33 PM, Devon McCormick wrote:
> >>
> >>> I used to have a named verb "idMat" in my own utilities library that
> was
> >>> defined this way:
> >>> idMat=: =@:i.
> >>>
> >>> but testing this on large arguments led me to formulate this
> >>> idMat=: 3 : '(2$y)$(>:y){.1' NB.* faster and scales better
> >>>
> >>> or, the same thing tacitly:
> >>> idMat=: (2 $ ]) $ 1 {.~ >:
> >>>
> >>>
> >>> On Tue, Nov 20, 2012 at 2:29 PM, Don & Cathy Kelly <d...@shaw.ca>
> wrote:
> >>>
> >>>> A key point that dawned on me, after reading all the helpful comments
> >> and
> >>>> suggestions that have come in, is that I don't need a verb- just a
> noun
> >> - so
> >>>> y=: 2 3 4 2 2 3 (it also works for an n,n array)
> >>>>
> >>>> ] u=:=i.# y
> >>>>
> >>>> 1 0 0 0 0 0
> >>>> 0 1 0 0 0 0
> >>>> 0 0 1 0 0 0
> >>>> 0 0 0 1 0 0
> >>>> 0 0 0 0 1 0
> >>>> 0 0 0 0 0 1
> >>>>
> >>>> and I can use u*y for different 'y''s ( as long as the # is the
> >> same)
> >>>> to get the desired diagonal elements with off diagonal elements 0
> >>>> I want this because I want to add to the diagonal elements of a
> matrix.
> >>>> This is for a Newton Raphson based power system load flow.
> >>>>
> >>>> Thanks to all,
> >>>> Don
> >>>>
> >>>>
> >>>>
> >>>> On 20/11/2012 8:24 AM, Roger Hui wrote:
> >>>>
> >>>>> http://www.jsoftware.com/**jwiki/Essays/Identity%20Matrix<
> >> http://www.jsoftware.com/jwiki/Essays/Identity%20Matrix>: 34 different
> >> ways
> >>>>> to generate the identity matrix of order n; the first is =@i. and the
> >> last
> >>>>> is _&q:@p:@i.
> >>>>> ------------------------------**------------------------------**
> >>>>> ----------
> >>>>> For information about J forums see
> >> http://www.jsoftware.com/**forums.htm<
> http://www.jsoftware.com/forums.htm>
> >>>>>
> >>>>>
> >>>>
> >>
> ------------------------------**------------------------------**----------
> >>>> For information about J forums see
> >> http://www.jsoftware.com/**forums.htm<
> 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
> >>
> >
> >
> >
> > --
> > 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
