stephenbrooks.orgForumMuon1GeneralBest250 and Bonus Particles
Username: Password:
Search site:
Subscribe to thread via RSS
Stephen Brooks
2002-08-14 17:02:24
Well we're almost at the 1% muon yield milestone, so as promised I have produced a 'seed' results.dat containing 250 of the best results from everyone in the project.  No user is allowed more than 10 entries in this (so we get a range), but other than that the best possible results are used in it.  The file is of course downloadable from the main Muon1 page.

As well as this, we've had some discussion about the fact that the more efficient designs take longer to simulate per particle, so the particle count isn't really the best way of scoring the results.  Instead of totally shaking up the stats (which would have annoyed some people), I've decided to make a kind of "bonus points" system for especially good results.

Muon yields 1% - 1.1999% score in the particle count as 5 times the number of particles actually simulated.  Yields from 1.2% - 1.3999% score as 6 times, 1.4% - 1.5999% as 7 times and so on.  This will mean a "particle" now reflects a unit of computation about equivalent to what a particle _was_ with the lower results around 0.3%.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
MaFi
2002-08-14 17:09:39
finally!
thats great, i was really waiting for that.

markus
Jet from ARS
2002-08-14 17:37:09
So, help out those of us who might be a little confused (ok, me):

After we download the best250.zip file, we unzip it, replace results.dat and restart the program?

What happens after we get 10 entries?  Do we have to get a new results.dat file?

Jet
thormodr
2002-08-14 17:47:40
You do one of two things...
You can replace your results.dat with the top 250 one
or
You can append the top 250 file to your own results.dat file...

The 10 thing that Steven was talking about was that only 10 results maximum ATTRIBUTED TO AN INDIVIDUAL USER (I WASN'T COMPLETELY CLEAR) will be in the downloadable top 250 file.  All you need to know is unzip and run it...

I hope that clears it up... big grin

[This message was edited by thormodr on 2002-Aug-15 at 16:27.]
Jet at ARS
2002-08-14 18:01:44
You bet.

Thanks!

JEt
TheFinalLoser
2002-08-15 06:11:42
whooo, this pack gave my computer a realy boost...
i get now results ten to 15 times higher in average then before...
but it takes surly quit long now...
Jwb52z
2002-08-15 09:22:14
I don't think Stephen meant that there were only 10 results in the Top 250 download dat file.  I think he meant that no more than 10 results from an indivisual user would be in it.
Pascal
2002-08-20 00:04:45
I'd like to discuss a bit the bonus system for good results. 
You get extra 'particles' for better results, that's quite ok. 

But in that way the real work that stands behind this (I mean the work for bad results up to 1%) does not appear real any more.  Right?  Think about getting results with percentages perhaps up to 3 %, that means a factor of 15 (3.000 to 3.1999 %)

So, wouldn't it be good to add the percentage of bonus particles in the stats?  I think thats not soo difficult to do.

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
Pascal
2002-08-20 00:04:52
I'd like to discuss a bit the bonus system for good results. 
You get extra 'particles' for better results, that's quite ok. 

But in that way the real work that stands behind this (I mean the work for bad results up to 1%) does not appear real any more.  Right?  Think about getting results with percentages perhaps up to 3 %, that means a factor of 15 (3.000 to 3.1999 %)

So, wouldn't it be good to add the percentage of bonus particles in the stats?  I think thats not soo difficult to do.

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
[ARS]odessit
2002-08-20 06:59:51
I actually do not mind this stats system at all.  While it sucks to be passed by a 3 day old cruncher (that was supposed to pass me in 6 days instead big grin) everyone has a fair chance in this.  The existance of Best250.zip is the proof.  I see not why it would hard to put % of bonus points into the stats, it would be redundant.

Crunch on more machines, update your results.dat and with just a little bit of luck you will be the king of this project if you want wink

Intel P3-M 866 MHz
Duron 1000 MHz
XP 1600+ @ 2000+
[Ars]kenlow
2002-08-20 10:37:42
[QUOTE]Originally posted by odessit:
I actually do not mind this stats system at all.  While it sucks to be passed by a 3 day old cruncher (that was supposed to pass me in 6 days instead big grin) everyone has a fair chance in this.  The existance of Best250.zip is the proof.  I see not why it would hard to put % of bonus points into the stats, it would be redundant.

QUOTE]

I think that the system is pretty fair too.  I mean it is the same system that is applied across the board.  Everybody benefits or suffer the same.  The only thing that may affect the results is people that constantly produce high yields will benefit from a very productive result.txt.  But Stephen has negated this by releasing the Top250 file.

The Top250 file is the result of people that have joined this project from the beginning; and some of us may have benefited from joining the project late and getting the benefits of the Top250. It is because of this that you see individuals with very high points vs results returned ratio.  It is because of the constant high yields acheived with the Top250 file.

odessit :- You were the one that told all of us at TAM to borg smile Well...I am up to 11 XP borgs now...and will be adding more next week big grin
Stephen Brooks
2002-08-28 08:53:52
Posts deleted about Pascal looking in the wrong "coloums". razz
Pascal
2002-09-04 09:06:30
How do you think about a new release of a TOP250?

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
Stephen Brooks
2002-09-04 11:00:22
Because of the inclusion of that manual result of mine, I'd like to let the results diversify a little before a new best250, but certainly I'll make one when/if it gets to 3%.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
Pascal
2002-09-05 10:07:09
Accordin' to MaFi's question what an effect will take placa by releasing a TOP... TOP5000, I (and also many others, as I know) would like to know something more about the client:

- if it only choses the best results - how many does it choose?
- what is the lowest yield / lowest percentage or is it e.g. the 100st lowest that it still chooses?
- so..all in all .. what results are used for 3 of the 4 genetic algorithms?
- how does the client detect any worse results... so is there any other yield computed, perhaps on the half way?  With that, the client even might choose some parameters from very low yields (at the end), that perhaps were higher at an ealier place. 

Remember, I know about the algorithm in general, as mentioned in your technical report up to june 2002.

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
gogomaus
2002-09-05 11:30:17
Pascal, what an inquisitive person you are.. wink
And what about opening a new thread; in case Stephen is willing to exhibit those sweet secrets...
Pascal
2002-09-05 11:46:51
quote:
Originally posted by gogomaus:
Pascal, what an inquisitive person you are.. wink
And what about opening a new thread; in case Stephen is willing to exhibit those sweet secrets...


Wolfgang, if I already were moderator of this board, I would be going to move these two or three postings to a new thread.  big grin

And if Stephen Brools talks about the algorithms, it's quite interesting to know a bit more about them, ain't it?

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
Pascal
2002-09-12 23:35:26
Now we have a new TOP250 and also a TOP1K.

what does the client do with 250 results and what with 1k results?  Where is there any difference?  How many, or which results does he look at?

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
Stephen Brooks
2002-09-13 10:09:04
Whenever I refer to an algorithm "selecting one of the better results in the database", I refer to a particular function fairlygood() in . This can actually select any of the results in the database, but is strongly biased towards choosing the higher-scoring ones.

To get an idea of what it does, generate a few random numbers between 0 and 1, and then raise them all to the power 0.1 (rand(0 to 1)^0.1).  You'll see a load of numbers come up - mostly around the high end of the [0,1] interval (about 0.9) but a few lower ones thrown in.  This is the sort of scheme that routine uses.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
Dave Peachey
2002-09-24 04:15:29
Stephen,

Having noticed that the top ten 'Best muon transfers' are all at or over 3%, I don't know whether it's in hand (albeit it's been noted in this thread) but maybe it's time to issue another Best 250 results file again?

I'd like a chance to get one of those high scores myself big grin

Dave
Stephen Brooks
2002-09-25 12:28:40
There are now new best250 and best1k files, going up to the current best 3.340% result.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
[Muon.FG]7F4
2002-09-26 01:44:10
I miss my top result in the results.dat (was sent 23.09 or earlier).  I guess, it's because the same muon-result (2.991501) like the result from [SG]Matai (the parameter are different).

I hope it's just an failure in the top250 creation code and you still have my result in your database.
Stephen Brooks
2002-09-26 03:39:40
Muon.FG results still suffer from the "nonstandard linefeed" problem, meaning my code that selects the best250 may not have picked up your result.  All linefeeds in your results.txt and .dat files ought to go as 0x0D followed by 0x0A and no other sequence of those two control characters (0A 0D 0D0D0A etc.) should appear.

So some or all of your results may not show up in the stats or the best250 - or for that matter Muon1 may delete them itself before sending because they are not recognised as being in the correct format.  For now, this is your problem, as on the whole it doesn't affect the database too seriously.  I may include a feature in a later version of Muon1 to 'fix' linefeeds that are incorrect.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
[Muon.FG]7F4
2002-09-26 06:48:15
linefeed problems for 7F4 and V3nd3tta couldn't exist, because I've checked

and in the teamstats the topresult was correct, but it doesn't matter, i merged the top250 and my top50, so it is in again.

PS.: take a look in your e-mail account (you find the linux-version of the muon commandline-client)
[ARS]odessit
2002-09-26 09:40:04
quote:
Originally posted by [Muon.FG]7F4:
PS.: take a look in your e-mail account (you find the linux-version of the muon commandline-client)


red face smile big grin wink razz eek cool
Now we are cooking!

Intel P3-M 866 MHz
Duron 1000 MHz
XP 1600+ @ 2000+
Stephen Brooks
2002-09-26 12:36:37
Take a look at the page now, the Linux client is there... smile
[ARS]odessit
2002-09-26 14:02:18
7F4 - Thanks!  big grin

Intel P3-M 866 MHz
Duron 1000 MHz
XP 1600+ @ 2000+
Pascal
2002-10-27 02:06:50
Is it possible to have a new TOP250 and a new TOP1000?

Many new participants come along, they may use this file results.dat for getting up in the stats wink

___________________________
Member of www.rechenkraft.net - German Website about Distributed Computing Projects

1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25163328 accesses.