We have moved down a rank

**PCZ** · 10-29-2003, 02:34 PM

The defection of Jeff from TPR to US Distributed improved there LifeMapper ranking to such an extent that it pushed them up to fifth overall.

DC Ranking

**rsbriggs** · 10-29-2003, 02:43 PM

Correct me if I'm wrong here, but isn't math2 our weakest area in the overall? I can put some boxen on SOB but I'm not sure it would help much... Anyone have any idea what WOULD help most with the overall score? More lifemapper?

**rshepard** · 10-29-2003, 03:14 PM

If it were me, I'd probably go for the PSCI category (maybe Climate prediction) or the MSCI category (maybe FAD) -- the gap in math 2 is so large we could just about concede it.
I'm pushing Lifemapper hard right now mostly because there's such a gap between our #11 spot and the #10 and above grouping. Once we "re-establish contact" there, I am probably going to split my boxes up something along the lines I listed.

**magnav0x** · 10-29-2003, 05:04 PM

I pressume we could get a bit off of UD (THINK), see how we are ranked 896 in the project.

**willy1** · 11-09-2003, 04:04 PM

We appear to have dropped another place in the DC Repository standings, to 7th.

Our composite score dropped almost a half point, to 5.59. That's a lot .

It's hard to tell where we lost ground - the only place I see and downward team movement is in DHEP, where we fell to 3rd. Rechenkraft.net moved up 5 places in that project in one day.

**magnav0x** · 11-09-2003, 05:02 PM

dhep IS the reason for the fall

**PCZ** · 11-09-2003, 05:31 PM

Best we get first place back again then

How much do we need ?

**magnav0x** · 11-09-2003, 05:54 PM

Sussex 37,858,911,638
Rechenkraft.net_Germany 8,860,085,098
Free-DC 5,493,117,928

A 333mhz computer does just over 1,000,000 a day. But this is a JAVA client

**ECL** · 11-09-2003, 05:54 PM

Did Rechenkraft move up due to huge output from their members, or did a bunch of people transfer their work?

To regain second place, we'd need to double our total work, which is probably a solid month of crunching for a dozen additional machines.

Has anybody got useful stats for this project, so we can see daily output and team moves?

**magnav0x** · 11-09-2003, 06:03 PM

The only stats we have right are for Free-DC only members in the history database @ http://stats.dbestern.net/dhep

I suppose I could do some that would show teams output as well. I'll talk to IB about getting a CSV for the teams and I'll get back to you guys.

**gopher_yarrowzoo** · 11-09-2003, 06:35 PM

Borg More Boxen... Borg More Boxen... Get those that have jumped ship to jump this way or jump back again...

oh incase ya missed it...

Borg More Boxen...

Borg More Boxen...

**magnav0x** · 11-09-2003, 06:43 PM

What was that second thing again?

**prokaryote** · 11-09-2003, 06:45 PM

Hi People,

There's some unique geometric properties to the DC Repository's mega stat that lead to some counter-intuitive optimization strategies (Due to using a Hamming type distance measure instead of a Euclidean measure). I'll run a quick vector analysis to see what mix of projects would give team Free-DC the most bang for the buck.

prok

**willy1** · 11-09-2003, 08:38 PM

What did he say?

**rshepard** · 11-09-2003, 08:44 PM

Translation: "Nobody completely understands how the DC Repository stats work, but I'll make a guess"

**PCZ** · 11-09-2003, 08:56 PM

So which statistic does DC rep use for dhep ?

The stats for that project are a bit confusing

Also we moved up several places in EON and GOT, that should have helped our overall ranking.

Also going from 11th to 5th in LifeMapper should have gained us some points.

**rshepard** · 11-09-2003, 09:03 PM

So which statistic does DC rep use for dhep ?

Looks like "Effort Units" is used

**magnav0x** · 11-09-2003, 09:10 PM

Unfortunatly the DC Repository rank is based on local percentage of the work that is done for the given project. Example (made up numbers):

Free-DC at first place in a project and has 25% LPW would yeild lower in their ranking than Free-DC at second place with 40% LPW. We WERE in second place in the DHEP project with 64% LPW, but now we are in third with 32% LPW. We did not loose so much ranking because we fell back one rank. We lost so many points because we lost the majority of our LPW for the project. LPW and the popularity of the project are what determine how much points you get towards your overall ranking. So our first place in DF does not get us a majority of our points. Also, noting that we only have 25% LPW in DF doesn't help much even though it is a relativly popular project (number of teams in the project). FAH2 and GAH would give better results (per LPW) in Biological Sciences, because of their great popularity.

**willy1** · 11-09-2003, 09:50 PM

So we're wasting our time with GIMPS, if we want to move up in the DC Repository, since ARS have in excess of 99% of the LPW.

We would want to work projects where we could get a largish percentage of the work, but not necessarily 1st place, and ignore projects where other teams have large unassailable LPW.

**Paratima** · 11-09-2003, 10:18 PM

It all sounds very much like bistro math to me.

**magnav0x** · 11-09-2003, 10:34 PM

You can say that running GIMPS is useless, because there is a large precentage of the LWP that we'll just never be able to achieve, but at the same time, some LWP is better than none. As you will see we now have 0.04 LWP for GIMPS. I think possibly running gimps for another month at least would be worthwhile. But yes, there are other project such as Seventeen or Bust, DPAD, and others that could greatly increase. In the week or two that I have been in DPAD our LWP has raised from 0.15 to 0.41. Mind you that is with only 2 or 3 active members. It also seems that once Free-DC approaches 3rd place in LM, that the possible LWP gains will become very minimum as the top 3 teams at the time will have about the same output. Also, with chessbrain in mind...even if US Distributed doesn't passes us, but maintains their current output, they will share a majority of the LWP with us...causing us even more troubles in the overall ranking. Pretty much, you just try to get as much LWP as possible! I currently have stuff on DPAD, Gimps, and CB. In about 1 or 2 weeks I'll pull out of Gimps, but I will remain on DPAD and CB, at that point I will help out on the DHEP front to try and regain some of our 30% LWP loss that we suffered yesterday.

Such projects where on team dominates and have say 60% of the projects LWP are probably the best ones to work on. UD for example: Team 2ch has 60% of the LWP. That is LWP that we can reasonably take from them. Projects where all the big teams have high output in are the ones that aren't worth the effort. When you see a #1 team with 25.3% or an the #2 team with say 25.1%, it's not work the effort. The LWP at that point is spread so thin that even if you climbed to the top you would then be spreading the LWP out thinner and would maybe achieve 18%. The most you can expet to do is get movement in the ranks, due to hurting the opposistion's possible LWP.

UD would also provide high results in the Medical Science rankings, because of the popularity level.

**prokaryote** · 11-10-2003, 12:44 AM

Okay sorry it took so long, I was waiting for a reply from Dave (the owner/creator of the DC Repository stats page) as to how exactly the ratings are calculated. The numbers that I was coming up with (based on how I was reading the formula in the FAQ section) didn't jive with the values displayed on the page until he set me straight on how he figures them.

The formula is as follows:

A team's rating for a project = {10 * [ (number of active (non-zero total) teams out of the 42 that are tracked) / (42 tracked teams) ] } * [ (Number of this team's WU's in this project) / (Sum of the 42 team's WU's for this project) ]

A team's rating for a category (for example, say the physical sciences category) is just the sum of that team's ratings in each project that make up the category.

A team's overall rating is just the sum of that team's rating for each category.

Hope that this clears up some issues brought up above.

Will get back with the analysis shortly.

**magnav0x** · 11-10-2003, 01:07 AM

so basicaly.....

popularity * LPW as I mentioned

Thanks for laying out the equation though, I couldn't get it right with the on in the FAQ either

**prokaryote** · 11-10-2003, 01:18 AM

Exactly.

Only, not as lop-sided or dominated by the huge team number projects, though still lop-sided.

There's also an added weighting caused by the geometry of how the stat is calculated that I'm working on (after I watch Discovery Channel Wings) that is counter-intuitive.

**willy1** · 11-10-2003, 01:48 PM

Originally posted by rshepard
Looks like "Effort Units" is used

Looking at the stats on the DHEP site "Most Work"

I can't see where Rechenkraft is anywhere near us in any metric

In the overall standings , by Effort Units:

Free-DC 5,526,249,422
Rechenkraft.net_Germany 822,006,756

It appears that if the DC Repository is using Effort Units, they are borked somehow.

**rshepard** · 11-10-2003, 01:59 PM

looks like maybe a glitch at the DHEP end-- I expect it'll get sorted on tomorrow's run at DC Repository

**PCZ** · 11-10-2003, 03:44 PM

So we are panicking over nothing ?

**willy1** · 11-10-2003, 04:16 PM

Without knowing the source for the DC Repository stats, it's hard to tell whether it's a DHEP or DCRepository problem.

Thread: We have moved down a rank

Thread Tools

Rate This Thread

Display

We have moved down a rank

Posting Permissions