stephenbrooks.orgForumMuon1GeneralDecayRotB - flat for over 6 months
Username: Password:
Search site:
Subscribe to thread via RSS
[TA]JonB
2007-03-26 12:54:07
The curve has been flat for many months, Stephen.  Any new optimizations planned?
Any tweaks you might suggest to find a way to make it jump like PhaseRotD just did?

[TA]JonB
K`Tetch
2007-03-26 19:21:57
well, tel you what, start playing with designs manually yourself jonB.  Thats how the design jumps often come in.  Or, if you're a samplefile user, disbale samplefiles for w while, especially if you've not got many highly clustered results.  Lets get a bit of variation going. 
RGtx
2007-03-26 19:36:40
K'Tech, Since November I have been running DecayRotB from scratch without samplefiles and have reached a Muon% of 0.343874, on Stephen's weekly plot of Muon% vs MPts, I am the very bottom breakaway trail.  After a few weeks of constant increments, progress has somewhat abated.  Looking at the design parameters one gets some idea of where progress could possibly be made but I'm not sure how to manually input designs, a little advice would be appreciated.
[TA]JonB
2007-03-27 02:04:09
K'Tetch, I'm in the same boat as RGtx.  I'm not sure how to manually input designs.  As you've said, there are 200 some odd parameters to deal with and I don't have a clue where to start.  I suppose I'll search through some old posts here for clues.

Do you use something like Excel or a database to help examine the old results?  Some sort of trending software?
[TA]JonB
2007-03-27 02:54:56
Now, a "quick look" through the data for PhaseRotD shows about 474 different variables.  While the data is formatted using semi-colons and can be imported into Excel with the hope of working with it, that is too many columns for Excel to handle.  Way too many.

I suppose I could just tweak the "d" values or just the "hd" or "hrf" values, but I am still feeling like I am trying to steer an elephant by grabbing its tail and pulling hard.  I don't have the tools I need for this.
Stephen Brooks
2007-03-30 11:20:05
Trivial answer: Excel 2007 has increased the number of columns from 256 to 65536 (!!)

I made a results2csv program available on the site, but I think you're right and it produces too many columns.  I've been worried a bit about the stoppage of these optimisations too, but am currently working on the code for a new one.  I might also try and "seed" these optimisations again to get them out of a stuck area.

Note that if you turn off Muon1's auto-view (V) in graphical mode and zoom in, you'll see the elements of the beamline labelled with names like S27, D27 in the decay channel and PRF8, PD8 in the phase rotator section.  Those are the initial characters of the optimisation variable names.  Then each element's parameters (voltage, etc.) are abbreviated and appended to the end, so prf8v is the voltage (V, normalised to 000-999) of the 8th RF cavity of the phase rotator (P).

I guess if you wanted rows instead of columns you could put a result in Word, replace the semi-colons by carriage returns, then import and make a 2-column Excel file for that one result. 
[TA]JonB
2007-03-30 21:05:13
I haven't upgraded to Excel 2007 yet, so I'll wait.  I have noticed the designations in the graphical view and thought they corresponded but couldn't quite decode the prefixes.  This will help.

RGtx
2007-03-30 22:34:16
Microsoft are providing a downloadable 60 day free trial of all their Office 2007 products.
[TA]JonB
2007-03-31 00:29:47
I'd be more tempted if I hadn't just installed Office 2003 Pro.  My company has a deal thru Microsoft and I got it for $23.95 (as long as I still employed).
So I finally upgraded from good old Office 97.
AMD-USR_JL
2007-04-01 17:55:47
Once you have tweaked it how would you tell Muon1 to run that simulation?
Stephen Brooks
2007-04-03 13:05:35
I'd really advise using a small script to split the results into whatever format you want (PHP or C would do it).  Even I don't use Excel 2007 (2003 Pro here)!  Probably wouldn't get on with that ribbon thing anyway.

--[Once you have tweaked it how would you tell Muon1 to run that simulation?]--

I've added a brief thing about this to the FAQ.
AMD-USR_JL
2007-04-04 23:22:23
Thanks Stephen, i got it to work, but now i have another question.  I used the highest DecayRotB score for my tweakings and after several nowhere near close muon percentages to the original, even after changing just one thing, i tried the original simulation.  It also did not get anywhere close to the .41, it got .392. Are the sample results the same exact ones as the originals?  Or is there something messed up with my computer?
Stephen Brooks
2007-04-10 14:53:29
No, there's a statistical spread.  Unfortunately it's rather hard to get rid of (and it is part of the reason why the scores are averaged if they are rechecked).  Fortunately the optimiser does roll uphill on average even with this spread, but it is an annoying feature.  I've got a few ideas on how to compensate for it but they're fairly complicated.
Stephen Brooks
2007-04-10 15:03:42
Er, anyone else noticed (as per the title of this thread) DecayRotB's not flat any more.  Is that the result of manual tweaking do you think?  You could compare the current top of the samples to your seeded ones to check for any resemblance.
RGtx
2007-04-11 08:59:13
"Is that the result of manual tweaking do you think?" Yes, I fine tuned my best design until I got a max of about 0.404%, my rather low MPt/run indicated I was losing too many particles earlier on, so I merely bolted on the first five components from the top result in your sample file.  Subsequent improvement is computer optimization (well most of it, anyway).
waffleironhead
2007-04-13 00:23:33
congratulations and thanks RGtx for breaking the deadlock on decayrotb.  >>>>>>now onto the progress!!!!!
Stephen Brooks
2007-04-20 14:28:49
It's really set the computer optimisation going - just look!



Sometimes it's funny how the manual tweaks do different things to the computer optimisation, so the two enhance each other.  This happened on an old one too.  Ideally, I'd be able to make the computer optimiser as creative as human fiddlers too, but doesn't seem to have happened yet.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25163557 accesses.