stephenbrooks.orgForumMuon1GeneralBest DC Project Poll
Username: Password:
Search site:
Subscribe to thread via RSS
HETTATLONGUN
2003-06-14 15:27:34
Heh, heh, heh!  Has anyone else voted?  Big Grin

http://koti.mbnet.fi/aquarius/links/link_distributedpower.html

The poll is just a little down on the right side of the page.  Wink

Ahh, morning....Nothing like a good cigar, a cup of coffee and the sound of hot shell casings hitting the deck....HETT[TEP]
Pollock[Romulus2]
2003-06-14 20:19:16
That thing would only let me vote once!  I want a recount.  Smile
[DPC]Stephan202
2003-06-15 01:27:46
Voted Big Grin

---
Dutch Power Cow.
MOOH!
Stephen Brooks
2003-06-15 11:46:04
ROFL.  Many more and we're going to be more popular than Distributed.net on that poll!

Today's weather in %region is Sunny/(null), max.  temperature #NAN°C
HETTATLONGUN
2003-06-16 13:16:32
Heh, heh, heh!  I just noticed:

Distributed.net (21) 3%
Distributed Particle Acceleration (25) 4%

Big Grin

Ahh, morning....Nothing like a good cigar, a cup of coffee and the sound of hot shell casings hitting the deck....HETT[TEP]
Stephen Brooks
2003-06-16 19:12:40
Back in 1999-2000 I used to run d.net on my machine.  I didn't ever reckon at that stage that a DC project of my own would be beating them in a poll (though it's not a comprehensive poll by any means).
I suppose I can't blame people for eventually getting bored of exhaustively checking 2^56 code keys, then 2^64 code keys, then 2^72...
Right now I have to say that the projects out of those that I think have most scientific value are the biomedical projects: protein folding, and the drug-discovery ones -particularly Find-a-Drug, though UD is also good.  D2OL I have serious reservations about.  I like SETI too - the fact that it's looking for aliens doesn't mean it's not good science - I think at the moment it's providing some good lower bounds on the expected distance to our nearest sentient neighbours.
I don't mention my project yet as I'll only _really_ be happy once I've gone a lot further with it and produced a presentable array of engineering results.  My target is probably next year's Neutrino Factory conference (NuFact'04) about this time next year, somewhere in Japan.  I'll be wanting to present a 15-20 minute talk there about this project and its results - my slides for that will probably go both on this website and the ISIS theory group page. I'll be working full-time on this work (with no other commitments) from Sept 1st so that gives a long window in which I can do things.

Today's weather in %region is Sunny/(null), max.  temperature #NAN°C
[DPC]Stephan202
2003-06-17 00:20:55
I'm curious about your thoughts concerning D2OL.
I have run that project too, and the science behind it seems reasonable to.  I know I'm not a scientist, but calculationg docking energies of known and directly available molecules sounds like a good idea.
Too bad though they have troubles from time to time with their servers.

---
Dutch Power Cow.
MOOH!
Stephen Brooks
2003-06-17 02:26:20
D2OL appear to use a genetic algorithm to calculate the optimal position (or 'conformer') for each molecule in turn.  While in principle there is nothing wrong with this, if you compare the rate at which it scans new molecules with the old UD or the THINK engine, you'll see that it is a lot slower.  There are two reasons for this - one is that conformer-fitting is quite a geometrical problem, so THINK uses a mathematical technique to vastly reduce the number of possibilities it needs to consider, without much reduction in the quality of the end result.  The other problem is that genetic algorithms are typically very inefficient in terms of computer time, and D2OL's one uses a very literal interpretation of the principle.  It really _does_ have a "population" of individuals and generations that these evolve through.  On each generation, at least 90% of the individuals 'die' without reproducing or being recorded, which is equivalent to throwing away 90% of your results.dat file in Muon.  This also means that within one generation (which can be thousands of 'designs', i.e. distortions of the molecule) there is no possibility for designs to improve upon each other, as they are all done independently.  By contrast Muon1's "do one at a time" technique is often quite far advanced by the time 1000 designs have been simulated on one computer (really the limiting factor for Muon1 right now is things-to-design, not the optimisation rate, although I have a few ideas how that too could be improved if necessary).

Anyway, the way they were doing it seemed not slightly, but _extremely_ inefficient, as if they didn't really know what they were doing.  But I guess that was just the impression I got from running their client and looking at what it seemed to be doing for a while.  So by no means is this a bulletproof argument against D2OL, I just personally treat them with a degree of skepticism.

Today's weather in %region is Sunny/(null), max.  temperature #NAN°C

[This message was edited by Stephen Brooks on 2003-Jun-17 at 12:12.]
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25163051 accesses.