Re: [Jprogramming] Verb definition + inadvertent train formation.

Don & Cathy Kelly Tue, 20 Nov 2012 18:54:36 -0800

I just ran this
   y=.i.9000
   (10)6!:2 '=i.# y'
0.178201
   (10)6!:2 '=i.9000'
0.178399

verb variations with &.>9000 are in the same ball park -much faster thanidMatt2 (0.511) but appreciably slower than idMatt 1 and 3



      idMatt1=: (2$])$1{.~>:           NB. Tacit and fast
    (100) 6!:2 'idMatt1 &.> 9000 '
0.063462

idMatt3=: ,~ $ 1{.~ >:
   (100) 6!:2 'idMatt3 &.> 9000 '
0.0643745


Don


On 20/11/2012 5:48 PM, bob therriault 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:
    idMatt3=: ,~ $ 1{.~ >:

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

Re: [Jprogramming] Verb definition + inadvertent train formation.

Reply via email to