stephenbrooks.orgForumMuon1GeneralLinac900Removable8c2
Username: Password:
Search site:
Subscribe to thread via RSS
[TA]Assimilator1
2010-01-09 10:55:09
Interesting how it flatlined for quite a while & then took off, & it's still climbing! 

Has this happened before?
yoyo
2010-01-10 08:12:03
I made an experiment.  For the Boinc users I do not use Stephens sample files any more.  I build my own.  I put the results in the sample files with the highest MPts value.  Maybe this influenced it.
yoyo
[TA]Assimilator1
2010-01-10 10:12:27
How do you build your own??
yoyo
2010-01-10 16:43:04
The sample files are just a number of results.  So I take all results from the Boinc user and put 100 with the highest mpts value into the samplefile.  Afterwards the new sample file is sent to the user.
[TN]opyrt
2010-01-11 09:39:23
Yoyo: I've been doing the same thing... Focusing on high Mpts.  But don't you think it would be better to split 50/50 between a high Mpts samplefile and Stephen's samplefile?

Assimilator1: For us who don't hava access to the results from all the BOINC users directly, we can pick results from the relevant graph to make our own samplefile.

If you click on different squares here: http://www.stephenbrooks.org/muon1/plots/graph1/?optim=Linac900Removable8c2
You will get sample results for Linac900Removable8c2.
Stephen Brooks
2010-01-11 15:15:30
I've wondered about using this 50/50 split in the main samplefile myself.  The problem is Muon1 will tend to select to highest muon percentage results in preference to the high Mpts one, so I don't know if a mixed samplefile would have the desired effect.
[TN]opyrt
2010-01-11 16:13:53
Just to clearify:
I was thinking Yoyo@Home would help this project even more if 50% of the clients were using the samplefile created by Yoyo (highest Mpts) and the other 50% were using the original samplefile from this site. 

That way, the lattices would be optimized in "two directions" at once.
Stephen Brooks
2010-01-11 16:19:05
Yes.  That would be better.  If I wanted to do this I'd have to detect the user's IP address (or some other identifier) to determine which one of the two samplefiles to give them.
Stephen Brooks
2010-01-11 16:30:39
Oh - I've just seen the effect of the BOINC client using totally different samplefiles - two totally different evolutionary pathways!  If you scroll down on the page [TN]opyrt linked to, you get a plot coloured by user and you can see the (currently blue) BOINC results following a different line.
RGtx
2010-01-11 18:24:08
"If I wanted to do this I'd have to detect the user's IP address (or some other identifier) to determine which one of the two samplefiles to give them."

The config file already contains the line:

Sample file URL (.bin allowed): http://stephenbrooks.org/muon1/samplefiles/{lattice}_100.bin

Surely, you could create a second sample file, say {lattice}_100(yoyo).bin, then those wishing to use the alternative set would simply amend their config text file.
Stephen Brooks
2010-01-11 19:58:25
Yes, I could make it voluntary instead.  Was kind of thinking about rolling it out for all users automatically if it improves the optimisation as much as it seems to.
[TN]opyrt
2010-01-11 23:08:56
I would say the most interresting part of this optimisation right now is around 1100Mpts 1.25-1.43%. It would be really interresting to see what happened if Yoyo@Home only used samples from that area for a day or two. 
tomaz
2010-01-12 07:55:52
Great Yoyo, it's good to see that theory works
http://www.stephenbrooks.org/forum/?thread=1336&bork=yldyichkba

Stephen, why don't recode Muon to treat all designs in samplefile equaly (without preference) ? 
yoyo
2010-01-20 22:10:26
I currently do not use any samplefiles from Stephen.  For samplefile generation I use only the results delivered by the Boinc users.  But all results are sent to Stephen and will be considered when samplefiles for the native client is generated.
I'm thinking if I should also use Stephens samplefiles as input for my samplefile generation, to extract the highest Mpts results from both.
But I will wait some more days to see how it evolves.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25161937 accesses.