tlolczyk

Do I really have to put anything here?
The title says it all.

mwotton

On Wed, 27 Aug 2003 11:48:05 -0500, TLOlczyk posted:

An undergraduate writing in C++. 346 lines, apparently, and took
full advantage of a large number of v. fast machines...

mrak

--
realise your life was only bait for a bigger fish
-- aesop rock

tlolczyk

I know Hudson claims to have won. He probably has, but f****ve me
for wanting proof other then someone claiming he has won.

007

Hi all

Can u tell me what this ICFP contest or what-ever it was, si all about??
Thanks

tlolczyk

And the excuse for C++ taking second prize?
( And even the Judges prize where the code was done in a combination
of C++ and Dylan. )

danwang74

gregm@cs.uwa.edu.au writes:


Um..yeah right and the majority of the "real world" programming problems are
much more suited to FPLs.

So what would be a good test of "programming". FPLs fit a niche. They fit
this nice very well, but the niche is still a small fraction of the "real world".

tlolczyk

First. The understanding of a languages's numerics had very little to
do with it. The numerics of every language could handle it with little
problem. Since most languages support bignums. ( In fact C++ might
be the nastiest if your platform does not support long long. )

Second. I would hardly call Dual 1.8 M Xeons very fast.
I would not be suprised if a profesional programmer had two [1 ] Dual
1.8M Antlons at home which would be comparable to Dual 1.8M PIVs.
The Xeons have a larger cache and therefore speed up the processing.
So the machines would be slightly faster than what you would see
in the home of a professional programmer ( or onhis desk at work ).
I would also contend with "large number" since that implies something
like a 100. A lot might be acceptable, but I would believe that a team
of four programmers might actually have more computing power (
including laptops and old machines ). BTW he could easily pick the
wrong approach and set his machines to solving the problem with an
inferior solution and watched. Then waste the rest of his time
watching.

Third. I saw several people who were unable to submit anything because
they were not able to reliably generate a trace ( including me ), so
they never even submitted a bad entry. It took quite some skill to get
that far. ( Alltough I had very rough descriptions in computer form,
which could have been polished and generate some traces. I knew these
traces would not be that good, and work to get better traces. )

Fourth. I believe that an optimal approach is to find some sort of
path which approximates the fastest path. A* will find you the
shortest path, but the shortest path may not be the fastest path,
so you need some sort of weighted A*. Once you have the approximate
path then you do an A* on the path ih phase space with very brutal
pruning. ( In the end every approach is really just doing A* on the
phase space, with some sort of pruing. The big question is what kind
of pruning works best? )

So it was a fair test of programming. I can't help but wonder if C++
won last years competition and Haskell this years, you wouldn't be
saying last year was a poor test and this year was a good one.

[1] I'm not being redundant. I mean 2 machines with dual processors.

artie gold

Could you elaborate?

--ag


--
Artie Gold -- Austin, Texas

tlolczyk

I would have liked to see the faces of the attendees.

:> Seriously, though, as cool as this year's problem was, it didn't
:> really test programming, just experience with heuristic search
:> techniques, access to computing power, and (to a small extent)
:> understanding of one's language's numerics.

: So what would be a good test of "programming".

I don't know if I could give you a "good" test in an absolute sense,
but I think some of the previous tasks have been "better" tests.

-Greg