Good discussion, thank you. More inline.
--- Roger Hui <[EMAIL PROTECTED]> wrote:
> a. What is "scan"? If it is f/\ and f/\. then
> there is little meaningful to be said in addition
> to what I have already said: If the interpreter
> "knows" that f is associative f/\ is fast, i.e.
> O(n); if not, f/\ is O(n^2). In contrast, f/\.
> is always O(n).
In this particular example, it is the narrow meaning
as it is called in APL [1]. However, thinking further
in terms of systematic guide, scan in wide sense would
also be some other operators that apply a verb in
an equivalent of a "for" loop, including infix and cut
in multi-dimentional sense. On the other hand almost
anything in J is equivalent to a "for" loop, so there
need to be a boundary to this wideness.
[1] C. Burke and R. Hui, J for the APL Programmer.
> b. What is "direct single pass scan" and what is
> "alternative multiple scan"?
By direct single pass I meant when elements are
traversed from first to last in one go (with memory
of last iteration). The closest is u/\
([^:(0=])) /\ 3 0 0 2 0 4
3 3 3 2 2 4
Ideally it would model this explicit efficiently
dirscan=: 1 : 0
z=. 1{.y
for_i. 1+i._1+#y do.
z=. z,((i-1){z) u i{y
end.
)
([^:(0=])) dirscan 3 0 0 2 0 4
3 3 3 2 2 4
Multiple scans in too wide sense here
means that there are mutiple for loops
applied to the argument, in this case
just primitives, not "scanning" operators.
({~ [EMAIL PROTECTED] >./\@:* 0~:]) 3 0 0 2 0 4
3 3 3 2 2 4
This scan pattern seems so simple in explicit code,
that it's surprizing that in J to run it
efficiently it requires quite a trick:
([^:(0=]))~ /\.&.|. 3 0 0 2 0 4
3 3 3 2 2 4
> c. If "this problem" is the one originally posed
> -- repeat a non-fill item until the next non-fill
> item -- then I disagree that it takes a lot of
> effort to figure out in J. Since you said that
> "this problem seems very trivial" in C, then
> perhaps I am entitled to tell you that I solved
> the problem instantly in J. (I mean that I got
> the crux of the solution instantly. Actually
> programming it would probably have taken 5
> minutes or so.)
Not all are so fortunate. How long will it take a
J beginner or even an enthusiast? I consider myself
fairly proficient in regex, but I consult guides
all the time. This is an arguments of this particular
aspect of programming in J as a mental capacity
or knowledge acquisition.
> d. If you mean by "that reference" "Appendix B:
> Special Code" of the dictionary, then I should
> point out that Appendix B has been mentioned
> just recently in this forum.
By reference I meant a link to the fictional
Guide to Scans in J. Speaking of Appendix B, it
is a good reference, but it does not have the
connecting the dots part, which would be in a guide.
> Anyway, if you care about efficiency, then
> Appendix B is a standard place to look.
> On the other hand, it is also reasonable to
> program in J without regard to Appendix B,
> and then tell us about it if your J program
> is slower than you expect. (Whence most likely
> we will clear up your misconceptions :-)
>
> Many the items in Appendix B came about when I
> do benchmarks on programs submitted to the Forum
> by people whom I respect, and they ran slower than
> I expect. A number of them are "JKT specials";
> perhaps in time there will be a few "OK specials".
There are limits to feasibility of specials,
to understand them and related aspets of scans,
and because it's better to have a link to read than
to ask questions every time, it's good to have a guide,
if such guide is possible.
> e. Pepe's tools are indeed cool. See also Sample
> Topic 19 in the dictionary:
> http://www.jsoftware.com/help/dictionary/samp19.htm
> You know, you really should read the documentation
> sometimes. :-)
That's an excellent example. But it's all scattered.
And our brains need some help to get it together, such
as a Wiki page with links to all such tools.
> ----- Original Message -----
> From: Oleg Kobchenko <[EMAIL PROTECTED]>
> Date: Wednesday, December 13, 2006 4:38 pm
> Subject: Re: [Jprogramming] What a drag
>
> This is understood. What I meant by systematic is
> that Scanning is one of fundamental properties of
> J programming, as in "loopless code". Also it is
> generalized from classic APL.
>
> So it would explore different types of scan as in
> Jose's examples and examine what properties and
> constraints they have. Such as when direct single pass
> scan would be inefficient compared with alternative
> multiple scans.
>
> This problem seems very trivial, at least for C code.
> Yet in J it requires a lot of effort to figure out.
> It would be good, next time, just to point to that
> reference.
>
> Is it possible, or it's like to ask for a program
> that does anything?
>
> I also liked Jose's tool to construct and visualize
> application of verbs to arguments. It would also be
> good to centralize a collection of various such tools
> in a toolbox in Wiki. ts is another such tool,
> Primitives to Mnemonics, etc.
>
>
> --- Roger Hui <[EMAIL PROTECTED]> wrote:
>
> > Appendix B of the dictionary
> > http://www.jsoftware.com/help/dictionary/special.htm
> > says that:
> >
> > +/\ monad also * and - ; on Windows, assembly code for integer
> > arguments
> > +/\. monad also * and - ; on Windows, assembly code for integer
> > arguments
> >
> >
> >
> > ----- Original Message -----
> > From: Oleg Kobchenko <[EMAIL PROTECTED]>
> > Date: Wednesday, December 13, 2006 1:32 pm
> > Subject: RE: [Jprogramming] What a drag
> >
> > > OK. So +/\ is special code then.
> > > Thanks, Raul and Jose.
> > >
> > > Q=: _2+?1000$5
> > > (+/\ -: ([+])/\) Q
> > > 1
> > > (+/\ -: ([+])~/\.&.|.) Q
> > > 1
> > > ts '+/\ Q'
> > > 1.02999e_5 5248
> > > ts '+/\.&.|. Q'
> > > 1.50313e_5 9728
> > > ts '([+])/\ Q'
> > > 0.311365 18304
> > > ts '([+])~/\.&.|. Q'
> > > 0.000485823 10496
> > >
> > >
> > > Still, it's kind of murky.
> > > Is there a link where it is systematically treated?
> > >
> > > Also here is a relationship,
> > >
> > > -/\1 2 3 4 5
> > > 1 _1 2 _2 3
> > > -/@|.\.&.|.1 2 3 4 5
> > > 1 _1 2 _2 3
> > >
> > > -/\.&.|.1 2 3 4 5
> > > 1 1 2 2 3
> > > -/@|.\1 2 3 4 5
> > > 1 1 2 2 3
____________________________________________________________________________________
Yahoo! Music Unlimited
Access over 1 million songs.
http://music.yahoo.com/unlimited
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
