I suggest you try the benchmarks in
http://www.jsoftware.com/jwiki/Essays/Index_in_Nub
in APL.
Another one in the same vein is /:/:y to find the
rankings of y . The grade on the left is often
avoided in APL because it takes n log n time whereas
the equivalent
i=. /: y
z=. (#y)$2
(i.#y) i}z
is linear in the second part. In APL it's
i{is}{gradeup}y
z{is}({rho}y){rho}2
z[i]{is}{iota}{rho}y
In J you just do /:/:y because grade on small
range integers takes linear time.
There is also the "special code" in J.
http://www.jsoftware.com/help/dictionary/special.htm
The J architecture makes it straightforward to
"recognize" tacit expressions and provide efficient
implementations of them. e.g.
y=: 1e6 [EMAIL PROTECTED] 1e6
5e5 (< i. 1:) y
3
Index of the first item of y greater than 5e5.
ts=: 6!:2 , 7!:[EMAIL PROTECTED]
ts '5e5 (<i.1:) y'
9.77778e_6 1216
ts '(5e5<y)i.1'
0.00713191 1.04934e6
In APL there is no recourse but to do (5e5<y)i.1 .
----- Original Message -----
From: [EMAIL PROTECTED]
Date: Friday, June 1, 2007 7:46 am
Subject: Re: [Jgeneral] Re: Challenge to expert J'ers
> OK> (/:@({~ i.~) -: tgu3) x
> 1
>
> I am in a state of shock !
> Roger ! I can't believe my eyes on the difference in speed of the
> i.
> implementation versus dyadic iota in APL why is it so ?
>
> I would qualify Oleg's solution as the "pure J" solution and it
> should be
> the "pure APL" solution too but...
>
> Oleg's solution can't be implemented in APL because it is about
> 323
> seconds with an x which is ten times smaller. No APL'er I know
> would have
> taken this simple approach because they are brain damaged by their
> APL
> implementation.
>
> History: Us APL'ers have been influenced by the quality of
> implementations
> of APL interpreters. It would have never crossed my mind to use
> dyadic
> iota (i. for J'ers) in such a straight forward manner. Why ?
> Because it is
> slow as a dog in all APL interpreters I currently know.
>
> I remember several conversations with various APL'ers about how
> some
> "motivated newbies" wanted to implement a "faster dyadic iota"
> rather than
> doing the "right thing" and move forward... Or was it really
> backward ?
>
> Conclusion: Even older APL'ers have to understand the influences
> of the
> implementations they worked on and they should throw all that
> knowledge
> away because J's implementation makes it really clear that the
> particulars
> of any/all APL interpreters have short-circuit their brain away
> from the
> simplicity of the language.
>
> NB. I am so shocked that I am preparing a white paper on this very
> issue
> for our APL team. The title could be:
> "How APL implementations have ruined your pure APL thinking and
> why J can
> restore it..."
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
