stephenbrooks.orgForumOtherBrowser Warssuggestion
Username: Password:
Search site:
Subscribe to thread via RSS
excaliber[Free-DC.org]
2003-12-08 14:25:30
Quick suggestion.  I generally keep tabs on when a new Best100 result file is added.  But wouldnt it be nice if the client automatically got the latest from one of the FTP servers?  That way everyone would be the most up to date on the latest optimizations.  I would think that would speed up optimizations, as there would be less people working on inefficient work units.

Just a thought.
Lex_Luthor
2003-12-08 22:12:55
I suppose a feature like this could be debated as a future option (default: off) much like the client auto-update has been talked about in the past, but of course many prefer not to append their results.dat in order to generate purely unique results.  If everyone downloaded and used the same results file as a basis for their optimisations, then the project would definately be limiting itself in how far it could develop.  Sure, everyone would get different results, but definately not as varied as could otherwise.  Important mutations could be lost.  Remember, the "best" results had to originally come from somewhere.  Think of the gene pool - in a situation such as this, we can clearly see what happens when the same data gets used over and over: you get characters from Deliverance. Eek I agree that it can be beneficial to add a "best" file to build off of, but from there I think it would be better to let your specific client come up with it's own unique results instead of appending to your results.dat so often.  At least for the sake of the project itself, and putting the stats list "contest" aside.  Wink

"We're sorry to bother you at such a time like this, Mrs.  Twice.  We would have come earlier, but your husband wasn't dead then."

[This message was edited by Lex_Luthor on 2003-Dec-09 at 6:33.]
[DPC]Stephan202
2003-12-08 22:57:27
Stephan202 is happy to see that Lex_Luthor is with him Smile

As for the stats contest: Since it's about MPTS, good or bad results barely make any difference in the stats.

---
Dutch Power Cow.
MOOH!
Lex_Luthor
2003-12-09 00:10:07
Very true- but I do still see a thread here and there where it seems people complain about low muon percentages even though they have been running the simulation longer than others, and therefore add every new "best" file which boosts this (and causing the "inbreeding" problem), thinking that it really means something.  Of course the percentage would matter more, if we all started from scratch and stayed independent from each other.  I think that this misconception may be due to the fact that the stats charts are sorted by default according to the muon percentages (of course, this is no longer the case for "all optimisations").  Without reading up on the project a bit, many users may be misled, and also for some reason may think that (without really knowing what MPTS are) that the "best" files are somehow boosting their rating in the stats.  Red Face At least this is the impression that I sometimes get- I could be completely wrong.  Just in case- anyone not clear on MPTS, here's a simple explanation by Stephen.

By the way- Stephan202- glad to be on your side as well Big Grin

"We're sorry to bother you at such a time like this, Mrs.  Twice.  We would have come earlier, but your husband wasn't dead then."
Stephen Brooks
2003-12-09 00:53:09
quote:
Originally posted by excaliber[Free-DC.org]:
But wouldnt it be nice if the client automatically got the latest from one of the FTP servers?
Strange you should mention this _now_. I added a feature to automatically download and merge a new sample file with your results.dat every N days to the development source of 4.33 yesterday!  At the moment it downloads it from stephenbrooks.org, though there's a URL in the config file you can change to point to (say) a team's script-generated file.  It also makes sure it doesn't add results that are already in results.dat.  The default is currently to get a new sample-100 every 3 days, though I could change this to defaulting to off.

The sample files were designed to be a bit less nasty than the best files from the point of view of inbreeding concerns: they do contain a lot of the highest results, but also random ones from all the way down the yield scale.

What would be useful to know at this point is where the new bunch of results that have pushed the SolenoidsTo15cm yield from 9.7 to 9.8% have come from.  If they are from a nearly-independent optimisation, then that would suggest defaultly downloading the sample files would be a bad idea.  If it just came from someone who already had the topmost result, then it's just the optimiser pressing forward, and it would seem better to exchange results more.
HB Pencils, also sold as "Moron's Choice" Graphite Cigars.
excaliber[Free-DC.org]
2003-12-09 04:11:28
Ahh, didnt think about the inbreeding problem.

So if I would clear out my results.dat right now, id still get the same kind of credit?

Also, would it be scientifically better if i did clear out my results.dat?  Would there be a better chance of evolving a better simulation, that we couldnt get with the current simulation
[DPC]Stephan202
2003-12-09 06:09:16
quote:
Originally posted by excaliber[Free-DC.org]:
So if I would clear out my results.dat right now, id still get the same kind of credit?

If you look at the mpts, yes.  If you want to show up with a very high mpts/result ratio, no.
quote:
Also, would it be scientifically better if i did clear out my results.dat?  Would there be a better chance of evolving a better simulation, that we couldnt get with the current simulation

Yes, though it's very unlikely that precisely your new breed will make it to the top.  It's about randomness and luck.  Scientifically it would - probably - be better.

quote:
Originally posted by Stephen Brooks:
What would be useful to know at this point is where the new bunch of results that have pushed the SolenoidsTo15cm yield from 9.7 to 9.8% have come from.  If they are from a nearly-independent optimisation, then that would suggest defaultly downloading the sample files would be a bad idea.  If it just came from someone who already had the topmost result, then it's just the optimiser pressing forward, and it would seem better to exchange results more.

That's an interesting thought.  I suppose you have the info needed to find that out.

---
Dutch Power Cow.
MOOH!
[DPC]TeamNWW - Huub
2003-12-09 07:41:38
quote:
Originally posted by Stephen Brooks:
What would be useful to know at this point is where the new bunch of results that have pushed the SolenoidsTo15cm yield from 9.7 to 9.8% have come from.  If they are from a nearly-independent optimisation, then that would suggest defaultly downloading the sample files would be a bad idea.  If it just came from someone who already had the topmost result, then it's just the optimiser pressing forward, and it would seem better to exchange results more.
HB Pencils, also sold as "Moron's Choice" Graphite Cigars.


I noticed the jump when i stated using 4.32G.. My results were almost immediately higher then before..
Stephen Brooks
2003-12-09 08:24:09
quote:
Originally posted by [DPC]TeamNWW - Huub:
I noticed the jump when i stated using 4.32G.. My results were almost immediately higher then before..
Well that worried me for a while because I thought that might mean the queueing on 4.32g was broken, and sure enough, virtually all the high results in the sample file are from 4.32g. However, there is a v4.32e one in there at 9.729984, which is higher than the plateau we were at before, so I think in fact this rise started before 4.32g, it's just that most people are using 4.32g now because it's the most stable version.

HB Pencils, also sold as "Moron's Choice" Graphite Cigars.
excaliber[Free-DC.org]
2003-12-09 11:33:58
Well, I think I may clear out my results.dat.  I dont really care about being on the best percentage list, and there are plenty of other users that can optimize the current branch.  Cant hurt to try a new one.  Wink
[DPC]Stephan202
2003-12-09 14:38:16
That's the spirit Smile

---
Dutch Power Cow.
MOOH!
excaliber[Free-DC.org]
2003-12-09 17:44:56
Smile
Lex_Luthor
2003-12-09 23:52:53
Good for you.  You're an inspiration to us all.

"We're sorry to bother you at such a time like this, Mrs.  Twice.  We would have come earlier, but your husband wasn't dead then."
rluce
2003-12-10 05:14:16
Hello Stephen

I noticed links to other stats pages in the message board.  Will these be permanent stats pages?  Will you be showing these on your main web page and links to them?

rluce
Stephen Brooks
2003-12-10 07:48:51
What sort of "other stats pages"?

HB Pencils, also sold as "Moron's Choice" Graphite Cigars.
rluce
2003-12-10 09:35:05
Hello Stephen

The stats i asked about are:

http://muon.dcstats.de/

http://stats.dbestern.net/muon1

rluce
excaliber[Free-DC.org]
2003-12-10 10:47:56
Whoo hoo!  Im at .302%!!  Yea!!

Quick question.  My results.txt is only 5kb, but my results.dat is over 100. Are the majority of those so bad that they never made it into the results.txt?

Thanks!

EDIT: Stupid me.  I have autosend on, so it sent the results in.  Thats why my results are low and my .dat high.  Silly me!  Smile
Herb[Romulus2]
2003-12-10 11:34:18
quote:
What would be useful to know at this point is where the new bunch of results that have pushed the SolenoidsTo15cm yield from 9.7 to 9.8% have come from.  If they are from a nearly-independent optimisation, then that would suggest defaultly downloading the sample files would be a bad idea.  If it just came from someone who already had the topmost result, then it's just the optimiser pressing forward, and it would seem better to exchange results more. 


I think I can give you the answer to this questionBig Grin
The source is this value:
9.671563 (1513.2 Mpts) [v4.32f beta] {125BF5E3} by [DPC]TeamNWW
it was way different to the family before, what found its maximum here:
9.680288 (1630.2 Mpts) [v4.32g] {B6263968} by [DK]Abyss

I've isolated that result a few weeks ago and Pollock and me raised numberous new rechecked values above the last family breed.  All the top 60 are only misc variations of this breed.

Last week, I found another family as of:
9.678355 (1504.9 Mpts) [v4.32f] {42077F65} by Lex_Luthor
which on spot produced:
9.707678 (1506.9 Mpts) [v4.32f] {48F6DB7C} Herb[Romulus2]
but unfortunately that was the end so far.

It's a matter to detect new breeds and work with them, to see if they continue growing.

You ask, how the heck, do I do that?
I use an spreadsheet with conditional formats and some statistical corner values versus a graphical trend.  The rest ist stomach Wink

The spreadsheet

-------------------------------
I'd say more, but I can't reach the keyboard from the floor.
Lex_Luthor
2003-12-10 13:23:04
quote:
Originally posted by excaliber[Free-DC.org]:
Whoo hoo!  Im at .302%!!  Yea!!

Quick question.  My results.txt is only 5kb, but my results.dat is over 100. Are the majority of those so bad that they never made it into the results.txt?



Your txt file is low because every time it reaches a certain size, it is sent in to the server.  These are only the current results that your machine has recently been working on.  Notice that after results are sent, you don't even have a results.txt in your Muon1 directory until the simulation has run enough to generate a new one.  Your results.dat file will continue to grow, as this file contains a compilation of past results, and the program uses these to work off of.  Big Grin

"We're sorry to bother you at such a time like this, Mrs.  Twice.  We would have come earlier, but your husband wasn't dead then."
excaliber[Free-DC.org]
2003-12-10 16:31:02
Yup, I just forgot.  I have cable now, so its set to autosend.  Last time I ran DPAD it was on dial-up, so i had autosend off.

.4% now!  Big Grin
excaliber[Free-DC.org]
2003-12-11 12:20:36
Woot!  Made 1.02% today!  Smile Im on a roll.  Too bad it slows down the processing the higher the percentage, so my percentage increase will only get slower.  Frown

http://intheory.ath.cx InTheory - Game Design and Idea discussion forum
AySz88
2003-12-11 17:32:51
quote:
I use an spreadsheet with conditional formats and some statistical corner values versus a graphical trend. 


Didn't someone make a ViewResults that plotted many parameters of results, where you could choose the parameters represented by each axis of a 3-d graph and the color of each point?
excaliber[Free-DC.org]
2003-12-11 17:51:01
That spreadsheet is quite impressive!  Too bad I have no idea what any of it is.  Is there a nice tutorial or guide anywhere that describes what each parameter is, and how to interepret the data?

http://intheory.ath.cx InTheory - Game Design and Idea discussion forum
Lex_Luthor
2003-12-11 19:42:45
quote:
Didn't someone make a ViewResults that plotted many parameters of results, where you could choose the parameters represented by each axis of a 3-d graph and the color of each point?


Are you referring to the Advanced View in View Results?

"We're sorry to bother you at such a time like this, Mrs.  Twice.  We would have come earlier, but your husband wasn't dead then."
Herb[Romulus2]
2003-12-12 01:50:57
Iwas afraid, someone would ask for a manual Big Grin there is none Eek

Copy the "values only" of an interesting line into the yellow first line and you'll see how all others compare to it.
The second line is for free playing around, I use it for queue layout.  Try it and see how colours are changing by different values.
3,4,5 and 6 are self explanatory.
7 is used as additional reference design for crosschecking
8 is a check function for current queue.txt values to see what will come out (needs special text to data conversion adjustment)

Todays version reflects a design trend I'd checked last week with no positive result, but I started one box again with it, as the current designs seem to be maxed out already.

When you look over the lines you see a few (marked the first entry in red letters) which behave somehow similar in layout.  These I've isolated and gave them as results.dat for one box to work with it.

All conditional format refers to the line 1 column and sets the background colour acc biggger or smaller values, equals are blank. 
The statistical lines show up in red if the value corresponds to the reference design.  This is just to get a feeling if standard statistical values play a role onto the parameters.

The colors in column 1 are handmade and highlight some thinks like my results, teammate results, special values etc. The source key for nearly all the current high values is highlighted in magenta.

Importing new data is easy.  Just fill in a new empty line where it belongs to.  Copy the first relevant parameter line from results.dat and put it into the empty column B cell.
Copy the reults line into the A cell.

Formatting the B-Cell into a complete dataline: Data > Text to Columns > Separated > Highlight all possible separators and fill in = for extra separators, and ready.
You need to do this only one time, all following pastes will convert automatically.

Clear as mud, eh?  Wink

-------------------------------
I'd say more, but I can't reach the keyboard from the floor.
excaliber[Free-DC.org]
2003-12-12 11:48:01
Well, I'll take another look at the spreadsheet and try to digest what you said.  Try that is...

Nother question.  I asked this last time i ran DPAD, but never got a good answer.  My progress is creeping along really slow now, and was wondering how manual optmization works.  How would I go about speeding optimization up a tad?

Thanks

http://intheory.ath.cx InTheory - Game Design and Idea discussion forum
Herb[Romulus2]
2003-12-12 16:16:40
Hehe, you just got nearly the whole procedure, now work it out yourself a bit Wink

-------------------------------
I'd say more, but I can't reach the keyboard from the floor.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io

Site has had 20647155 accesses.