Paul,

You have correctly identified the fact that (float i. float) is not
special-cased in Dyalog APL (unless comparison tolerance is zero, in which
case we DO hash it); we do a linear search and that is slow when the arrays
get big. A public bashing on the J forum still isn't quite enough to get the
Dyalog development team fired up about this, mostly because our imagination
fails us as to what sort of (sensible) applications there are for fast fuzzy
searching of floats.

If I got this right, your need for it actually arises from the fact that
neither APL nor J take comparison tolerance into account when sorting
floating-point values, probably for the same reason (cost seemed to outweigh
the benefits).

I think your comment that we have thereby contributed to rotting the brains
of generations of APLers is *slightly* over the top :-). I think Olegs
elegant solution WOULD have occurred to many APL programmers, in fact I seem
to remember having seen this as a puzzle solution a decade or two ago.

Congratulations to Roger on his excellent implementation! If someone comes
up with a convincing set of examples to motivate further work on this case,
we may change our minds (but please don't use the J forums for this). We ARE
doing a significant amount of work on optimizing various cases in a number
of primitives in the next release of Dyalog APL.

Morten

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] 
Sent: 6. juni 2007 17:48
To: General forum
Subject: Re: [Jgeneral] Re: Challenge to expert J'ers

Actually Roger was not aware of the definite problem with Dyalog APL's 
f{iota}f implementation which was highlighted in later posting. So his 
sentence would now probably contain more nuances.


Paul Gauthier
APL Software Developer - Senior
[EMAIL PROTECTED]
Phone: 312-739-3467
Fax: 312-739-3496

CheckFree. The Company that Powers Payment on the WebSM.
http://www.checkfree.com/paybillsonline




Brian Schott <[EMAIL PROTECTED]> 
Sent by: [EMAIL PROTECTED]
06/06/2007 07:10 AM
Please respond to
General forum <[email protected]>


To
General forum <[email protected]>
cc

Subject
Re: [Jgeneral] Re: Challenge to expert J'ers






                 Sorry to the forum and Roger.
                 I found the referrent finally, and I suppose Roger
is saying that that phrase was not the correct explanation.


On Wed, 6 Jun 2007, Brian Schott wrote:

+ Roger,
+
+                I cannot understand this post. First, I cannot find
+ the *previous* message to which you refer. Second, I cannot
+ tell what aspect of the previous excerpt you are "Not"ing.
+
+
+ On Fri, 1 Jun 2007, Roger Hui wrote:
+
+ + > In J you just do  /:/:y  because grade on small
+ + > range integers takes linear time.
+ +
+ + Not, because it is the straightforward thing and
+ + your faithful implementor has taken care that you
+ + don't pay a penalty for it.
+
+

(B=) <----------my "sig"

Brian Schott
Atlanta, GA, USA
schott DOT bee are eye eh en AT gee em ae eye el DOT com
http://schott.selfip.net/~brian/
----------------------------------------------------------------------
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

Reply via email to