stephenbrooks.orgForumMuon1GeneralIs it possible to avoid duplicates?
Username: Password:
Search site:
Subscribe to thread via RSS
[TN]opyrt
2009-01-04 00:23:30
I'm not sure my assumptions are 100% correct, so please bear with me if I'm wrong.

Sometimes we see that a lot of users find the same result.  And usually this result is found several times by the most active users.  If I understand correctly, the client desides what simulation to run based on a mixture of the latticefiles, samplefiles and datafiles and randomness... ?

If that is correct, wouldn't it be possible to make the client _not_ run a simulation if it already exists in the downloaded latticefiles?
tomaz
2009-01-04 01:40:34
Yes, but same results (muon%) doesn't neccessary mean same setup (same accelerator design).  And basically muon% is not result, result of our effort is design (parameters setup) or how should I call it...

My idea of optimizing CPU use is the way how queued designs are tested.  Often, first try is high and other are lower so design is unneccesary queued and computed 5 times.  Would it be possible to have such algorithm that first variation is always the lowest (worst) ?  That would reduce number of queued designs.
[TN]opyrt
2009-01-04 02:07:31
I thought all the parameters mentioned in the samplefiles were the actual parameters for the simulation... You know, where it says "#runs=5;#gen=6;d1l=999;d2l=999;d3l=999;" and so on...
I meant to use these as a check to see if a simulation has already been run.  It won't remove all dupes, but it will remove alot, won't it?
tomaz
2009-01-04 02:59:41
I don't know excatly how many parameters design has but let's say 200 or so.  Each can have value between 0 and 999. I didn't like probability math in school but as I can recall it should yield 1000^200 combinations (unique designs) or so... So there is very very slim chance to get two equal designs. 

Since there is so many possible designs and all can't be checked genetic algorithm is used.  That means that only "most promising" "breeds" are checked and further developed.  That's why we have samplefiles.  Designs from samplefiles are further "developed" (parameters adjusted...) and checked. 

The best results are than rechecked (queued).  There some variations in parameters are made to see how "stable" solution is.

Or something like that I hope that this is what you wanted to know.
Stephen Brooks
2009-01-06 17:13:22
Muon1 excludes identical *genomes* i.e. when the entire parameter list is identical to one already in the user's results.dat.  It does not check for identical runs between different users (so often several people will independently run the same one).

Also, with 1000^200 different designs there are not that many different scores to 6 decimal places (only a few millions), so some different designs will produce identical scores.

What we could have is several different people running the top simulation in the sample file, plus some small variations on it that do not affect the yield (no particle is lost or saved) but still count as different designs.

I suppose if you wanted to completely exclude these from your investigations you could remove all records in results.dat that have a score identical to one in the sample file.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25163220 accesses.