stephenbrooks.orgForumMuon1GeneralMailing list for version 5
Username: Password:
Page 1 2
Search site:
Subscribe to thread via RSS
CADkindaGUY
2003-04-09 14:21:00
--------------------------------------------------------------------------------
v4.3 works really fine!  Upload seems also to be working but I only get a funny message:

"New version v4.21b is available from the website!  (You are using v4.3)
Programm now reads from previous results properly and sending while offline is corrected.

*drivelingformuon1viewresults.exe*
----------------------------------------

I'm getting the same popup when it autosends results.
Stephen Brooks
2003-04-09 15:11:19
Oops.  Well I've just realised I need to get a file from work (the security module, so not on the webdrive) in order to recompile and fix this error.  For now I think I'll try uploading a new version-signal manually that will hopefully correct this bug.
-----
OK, that's fixed now, although the error box will now say "v4.3 download available from website!" so I think I better link up the HTML page to the new version (which I can now do as stats updates have temporarily been stopped).

Today's weather in %region is Sunny/(null), max.  temperature -99999°C
TheFinalLoser
2003-04-10 01:25:44
when do the new stats(v4.3) appear?

"So the last will be first, and the first will be last."
Mt.20:16
Stephen Brooks
2003-04-10 13:43:13
Bah.  I was nearly there with the new stats generator when a completely inexplicable (and rather untracable) memory bug appeared.  I'll try again this weekend and hopefully will get it running again.
-----
On further thought, I suspect this is a buffer over-run error, where someone's name caused a string to go outside the 100 bytes I allocated for it.  That sort of thing has unlimited capacity to arse up future events.

Today's weather in %region is Sunny/(null), max.  temperature -99999°C

[This message was edited by Stephen Brooks on 2003-Apr-11 at 9:02.]
Bill[Romulus2]
2003-04-10 17:43:13
Does this mean I didn't really submit my last results 113755 days ago?  Big Grin
Stephen Brooks
2003-04-11 03:23:27
quote:
Originally posted by CADkindaGUY:
*drivelingformuon1viewresults.exe*


You can download it from the page now.  No readme to tell you where to put it (i.e. inside the Muon1 install dir where results.dat is) as yet - I'll update that in v4.31 when I package the two together in the same download.

Today's weather in %region is Sunny/(null), max.  temperature -99999°C
[DPC] White Panther
2003-04-11 15:42:50
until when can you upload results from version 4.21b?

mooh mooh
Stephen Brooks
2003-04-12 03:49:24
I'm accepting both for now (all the way back to v4.0x in fact).  I can update both sets of statistics for quite a while to come.

Today's weather in %region is Sunny/(null), max.  temperature -99999°C
David
2003-04-15 08:13:26
quote:
Originally posted by Stephen Brooks:
If a design has a "true" percentage higher than the current maximum so far, then there will be a greater than 50% chance that its first run will turn up a percentage higher than that too, so it will be rechecked more than half the time.



This is true only if the distribution is symmetrical, surely?  I wonder how good this assumption is for low probabilities.
Stephen Brooks
2003-04-15 11:21:02
The distributions of interest here are all very-nearly normal, so it is symmetrical enough.  When the percentage transfer (p) is low, the distribution Binomial(n,p) (n=number of particles) can't be approximated to a normal curve so well, but I don't really care about low-transfer designs so much.  If you still don't believe that my method works, I did a numerical experiment with Binomial(20000,p) for various p and got it to print out the numbers of trials that came out above and below the expected mean (=20000*p) for 1000 trials at each percentage.
n=20000 p=0.500000: 493 below 501 above
n=20000 p=0.200000: 538 below 462 above
n=20000 p=0.100000: 490 below 510 above
n=20000 p=0.050000: 485 below 515 above
n=20000 p=0.020000: 523 below 477 above
n=20000 p=0.010000: 474 below 526 above
n=20000 p=0.005000: 512 below 488 above
n=20000 p=0.002000: 528 below 472 above
n=20000 p=0.001000: 548 below 452 above
n=20000 p=0.000500: 595 below 405 above
n=20000 p=0.000200: 647 below 353 above
n=20000 p=0.000100: 675 below 325 above
n=20000 p=0.000050: 733 below 267 above
n=20000 p=0.000020: 657 below 343 above
n=20000 p=0.000010: 822 below 178 above
This shows that it's near enough 50-50 down to p=0.001 (i.e. 0.1% transfer), and then below that you get more like two-thirds falling below the expected mean.  If I was serious about getting accurate percentages below the 0.1% level, I'd need a lot more particles and the optimisation would be that much slower.  For comparison the best designs are above 5% now.

Today's weather in %region is Sunny/(null), max.  temperature -99999°C
Bill[Romulus2]
2003-04-15 16:01:18
They're above 5% already?  Sheesh, I haven't exceeded 2% yet.  I need a smarter computer Smile
Flumsi
2003-04-15 19:30:02
quote:
Originally posted by Bill[Romulus2]:
They're above 5% already?  Sheesh, I haven't exceeded 2% yet.  I need a smarter computer Smile


Yes i have one at 5.69%. You dont need a smarter Computer, you need a better results.dat File.

@Stephen Upload the Best 250 Results please and Start Statistic Smile

Greez Flumsi
David
2003-04-16 15:51:07
quote:
Originally posted by Stephen Brooks:
The distributions of interest here are all very-nearly normal, so it is symmetrical enough.  When the percentage transfer (p) is low, the distribution Binomial(n,p) (n=number of particles) can't be approximated to a normal curve so well, but I don't really care about low-transfer designs so much.  If you still don't believe that my method works, I did a numerical experiment with Binomial(20000,p) for various p and got it to print out the numbers of trials that came out above and below the expected mean (=20000*p) for 1000 trials at each percentage.[code]n=20000 p=0.500000: 493 below 501 above
[snip data]



Oh, I wasn't disbelieving, really - I had been a bit concerned that the new method might slow the rate at which the initial climb occurred, but both your figures (snipped from the above) and a bit of thought makes me realise that this thought was bogus.
Stephen Brooks
2003-04-16 16:05:27
quote:
Originally posted by David:
Oh, I wasn't disbelieving, really - I had been a bit concerned that the new method might slow the rate at which the initial climb occurred, but both your figures (snipped from the above) and a bit of thought makes me realise that this thought was bogus.


That's OK.  I only "overreacted" with a numerical experiment because I've had trouble with statisticians arguing about this project in the past, and there's a certain sort that won't go away unless you give them some sort of scientific proof...!

The initial climb is an interesting phase - really it needs a different sort of scoring to the rest.  If I were to start a really _difficult_ optimisation where initially no particles at all would get through the whole thing 99.9999999999% of the time, a better scoring would be something to do with how far the furthest particles got.  Then if you have a _few_ particles (much less than sqrt(number in simulation orignally)) getting through, the statistics aren't all that nice and really I'd want a sort of trade-off from the distance-scoring to my current system.  But I'll burn those bridges when I come to them (or something Red Face ).

Today's weather in %region is Sunny/(null), max.  temperature -99999°C
ZeonX[OCAU]
2003-04-21 18:02:25
this thread has sorta changed from the original topic but can i be on the mailing list.  Could you use muon@mega-reviews.com.  Thanks.
BOHICASETI
2003-04-22 09:26:47
I would like to be tagged for the mailing list also please.  Thanks Cool

Friends help friends move....
REAL friends help friends move dead bodies.
ZeRoC00L
2003-06-19 12:58:54
As 4.31c is at the max.  (13,xx %), any news about a newer version ??
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25162677 accesses.