From: "Mike Bindschadler" <mikebind@seas.rochester.edu>
Path: news.mathworks.com!newsfeed-00.mathworks.com!webx
Newsgroups: comp.soft-sys.matlab
Subject: Re: entry-length
Message-ID: <eeefd44.121@webx.raydaftYaTP>
Date: Wed, 10 Nov 2004 18:24:12 -0500
References: <eeefd44.114@webx.raydaftYaTP> <eeefd44.116@webx.raydaftYaTP> <eeefd44.117@webx.raydaftYaTP> <eeefd44.118@webx.raydaftYaTP> <eeefd44.120@webx.raydaftYaTP>
Lines: 47
NNTP-Posting-Host: 66.67.50.109
MIME-Version: 1.0
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
Xref: news.mathworks.com comp.soft-sys.matlab:240966


Niilo Sirola wrote:

I just want to say that I like all Niilo's ideas. I think he's right
that the milestone prizes get new ideas into the open before the last
minute, and that's a good thing.

Niilo Sirola wrote:
>
> I actually went through the trouble of extracting the five
> "elementary solvers" from the leading entry (there were
> additionally
> several "meta-solvers" that used one or more elementary solvers
> with
> different parameters and random seeds) and running some benchmarks
> with individual solvers to find the optimal performance/time
> balance
> for different-sized problems, but I did not have much time to put
> in
> it and this approach sadly did not survive the competition however
> nicely ir was structured.

I had a similar idea, but never got to even attempting it. I'm sorry
to hear it didn't work out!
   My contest strategy was to try to get a working entry done in
Darkness, then see at the beginning of twilight how it compared to
other entries. This time, it was tantalizingly close to the the
leader, and this provided tremendous motivation for me to find ways
to make it better. I was able to do this, and won the twilight
phase. My first contest was the Gerrymandering contest, and I
learned in that one that once Daylight hits, the contest changes
dramatically, and the identity of the leader changes often as people
make incremental improvements on the current leading entry. I knew
that I wasn't going to have time to stay competitive after twilight,
so I figured that part of my contribution to the contest would be to
make my code reasonably intelligible so that people modifying it had
a chance of understanding the overall plan. For the rest of the
contest I have watched with interest and tried (mostly
unsuccessfully) to understand the pieces of the leading entries. If
I had had a compelling new idea about how to approach the problem
differently, I would have tried to develop it into a new entry, but I
didn't, and just watched.
 I'd also like to say that I really appreciated the descriptions
people posted of their algorithms. I found them very interesting, and
something I found it impossible to extract from the entries
themselves.
Thanks for a great contest!
-Mike