Statistics and scoring

kar_bon · 2008-01-27, 10:03

This thread contains information about the scoring system at NPLB.

On www.rieselprime.de -> menu "NPLB"
You can find many project information and links to the current scoring tables.
Also available is the status of the Team Drives with current reservations and completions and found primes too.

Scoring for a prime with exponent n:

$score = \frac{ln(n)^3*ln(ln(n))}{ln(333333)^3*ln(ln(333333))}$

Notes:

Every prime counts.
The divisor normalize the score against a prime of rank 5000 at Top5000.
If 260k < n < 333.2k then n = n + 70000 to make low ranges more attractive
so a prime at n=260k got the same score as a prime at n=330k.
Confirmation of already known primes scores half of a new found prime.
Example: new prime at n=340000 ⇒ score=1.0052939 and confirmation score=0.5026469.

Scoring for completed ranges:

$score = \frac{av * av * ln(av) * range}{10^{14}}$ where av is the average of the range ⇒ $\frac{range_{start}\,+\,range_{end}}{2}$ .

Notes:

If 260k < range < 333.2k then av = av + 70000 to make low ranges more attractive.
Example: The range from 340.0k to 340.2k scores 2.942925.

Scoring for LLRnet:

The score is gained from the k/n-pairs done.

Open questions:
none

Karsten

Flatlander · 2008-01-28, 15:14

Quote:

Originally Posted by kar_bon

Scoring like the Top5000 pages with:

log(n)^3*log(log(n))
-------------------------------- = score
log(333333)^3*log(log(333333))

means: the score of a prime at n is normalized by the score of the prime at place 5000 (nowadays n=333333).

The differences from Top5000 scoring:
- Every found prime wil count
- an already known prime verified count halve the score

Sounds fine to me.

gd_barnes · 2008-01-28, 20:09

I'm not fond of TOO many scoring statistics because they become unwieldly to maintain. I like the primes scoring approach but I'm not sure about having additional scoring for ranges searched or other things.

Karsten, you're call there. I think you said you would keep the scoring so I'll leave it up to you as to what you think you can maintain.

One thing I suppose we could do...add a little extra scoring for the lower ranges in drive #2 since there is somewhat less motivation to search them. Perhaps add ~70k the n-value for them for scoring purposes. This would make the 260k-333.2k range score like ~330k-403.2k. So people would actually get MORE score per CPU minute/hour/day since they take less time to LLR.

Another option is to score the drives separately but I feel we'd be getting into a little too much scoring and I don't think the incentive would be as great to score high on 'small drive' #2 compared to scoring high on an overall score for both drives.

Gary

em99010pepe · 2008-01-31, 22:36

I belong to a distributed computing team called Free-DC (www.free-dc.org). Like Riesel Base 5 PRP I would like to see NPLB with team support.

gd_barnes · 2008-01-31, 22:58

Quote:

Originally Posted by em99010pepe

I belong to a distributed computing team called Free-DC (www.free-dc.org). Like Riesel Base 5 PRP I would like to see NPLB with team support.

I figured it was something like that but I don't understand the context of the question in this thread. Are you asking Karsten or Anon to create a team for you?

Pardon my ignorance of these matters...

em99010pepe · 2008-01-31, 23:03

Quote:

Originally Posted by gd_barnes

I figured it was something like that but I don't understand the context of the question in this thread. Are you asking Karsten or Anon to create a team for you?

Pardon my ignorance of these matters...

I want him to add a column with the team name for each member on NPLB and process an overall stats page for teams..lol

mdettweiler · 2008-02-01, 06:53

I was just looking at the scoring page a moment ago, when I noticed that I was ranked right below CedricVonck for completed ranges--I had a slightly lower score (even though we each have done one range total so far), despite the fact that he had less candidates in his range than I did. And even though mine was a 2nd drive range (whereas his was a 1st drive range, and thus higher n), wouldn't it still be scored higher, because of the added bonus for searching non-top-5000 ranges?

kar_bon · 2008-02-01, 09:47

Quote:

Originally Posted by Anonymous

I was just looking at the scoring page a moment ago, when I noticed that I was ranked right below CedricVonck for completed ranges--I had a slightly lower score (even though we each have done one range total so far), despite the fact that he had less candidates in his range than I did. And even though mine was a 2nd drive range (whereas his was a 1st drive range, and thus higher n), wouldn't it still be scored higher, because of the added bonus for searching non-top-5000 ranges?

look a the stats page explanations:
your range was 260800 to 261000 (with bonus of +70000 -> 330800 to 331000). Cedrick got range 333400 to 333600.

calculation:
anon: 330900*330900*ln(330900)*200/10^14
Cedrick: 333500*333500*ln(333500)*200/10^14

so because your range is slightly lower (although bonus included) than the other and therefore the scoring is correct so far. a range from team dirve #2 of 263400 to 263600 is the same as that from Cedrick!
the number of candidates is not relevant here.

karsten

kar_bon · 2008-02-02, 18:14

if noone has anything against it, i will clear this thread next time.

post #1 rewritten with current scoring conditions and also question of teams.

jasong · 2008-02-03, 04:39

I can't help but note that after the Twin Prime Search primes fall off the list a few months from now, a person could find themselves in the position of actively crunching the project, maybe even increasing in rank, and yet getting lower and lower scores.

Maybe we should archive a list, starting now, of the digit length of the prime at 5,000th place every first of the month. If someone gets discouraged, they have the option of making new comparisons based on a non-moving target, the value of the 5,000th placed prime in a given month.

Is it possible to simply calculate a simple normalization value? (Not sure if normalization is the right term.

)

kar_bon · 2008-02-03, 05:54

i'm calculating all scoring in Ex*el and the divisor of the formula is stored as constant in one cell i use. so i only have to edit '333333' into '350000' (or what ever) and the new calculations will be shown immedeatly!
so, no problem.
or do you meant a new formula at all?

2008-02-02, 18:14	#9
kar_bon Mar 2006 Germany 3×23×43 Posts	if noone has anything against it, i will clear this thread next time. post #1 rewritten with current scoring conditions and also question of teams. Last fiddled with by kar_bon on 2008-02-02 at 18:14

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
ECPP - Scoring, or other primality tests (PFGW?)	f1pokerspeed	FactorDB	13	2012-07-02 09:04
Discussion about scoring results/primes	gd_barnes	No Prime Left Behind	33	2009-01-19 14:50
You know what they say about statistics ...	petrw1	PrimeNet	1	2007-10-08 13:29
321 Statistics	paulunderwood	3*2^n-1 Search	1	2005-02-25 21:41
Statistics	R.D. Silverman	NFSNET Discussion	1	2004-06-14 18:40

2008-01-27, 10:03	#1
kar_bon Mar 2006 Germany B97₁₆ Posts	Statistics and scoring This thread contains information about the scoring system at NPLB. On www.rieselprime.de -> menu "NPLB" You can find many project information and links to the current scoring tables. Also available is the status of the Team Drives with current reservations and completions and found primes too. Scoring for a prime with exponent n: $score = \frac{ln(n)^3ln(ln(n))}{ln(333333)^3ln(ln(333333))}$ Notes: Every prime counts. The divisor normalize the score against a prime of rank 5000 at Top5000. If 260k < n < 333.2k then n = n + 70000 to make low ranges more attractive so a prime at n=260k got the same score as a prime at n=330k. Confirmation of already known primes scores half of a new found prime. Example: new prime at n=340000 ⇒ score=1.0052939 and confirmation score=0.5026469. Scoring for completed ranges: $score = \frac{av * av * ln(av) * range}{10^{14}}$ where av is the average of the range ⇒ $\frac{range_{start}\,+\,range_{end}}{2}$ . Notes: If 260k < range < 333.2k then av = av + 70000 to make low ranges more attractive. Example: The range from 340.0k to 340.2k scores 2.942925. Scoring for LLRnet: The score is gained from the k/n-pairs done. Open questions: none Karsten Last fiddled with by kar_bon on 2010-02-25 at 01:42 Reason: Menu selection changed

2008-01-28, 20:09	#3
gd_barnes May 2007 Kansas; USA 2×5×1,109 Posts	I'm not fond of TOO many scoring statistics because they become unwieldly to maintain. I like the primes scoring approach but I'm not sure about having additional scoring for ranges searched or other things. Karsten, you're call there. I think you said you would keep the scoring so I'll leave it up to you as to what you think you can maintain. One thing I suppose we could do...add a little extra scoring for the lower ranges in drive #2 since there is somewhat less motivation to search them. Perhaps add ~70k the n-value for them for scoring purposes. This would make the 260k-333.2k range score like ~330k-403.2k. So people would actually get MORE score per CPU minute/hour/day since they take less time to LLR. Another option is to score the drives separately but I feel we'd be getting into a little too much scoring and I don't think the incentive would be as great to score high on 'small drive' #2 compared to scoring high on an overall score for both drives. Gary

2008-01-31, 22:36	#4
em99010pepe Sep 2004 2·5·283 Posts	I belong to a distributed computing team called Free-DC (www.free-dc.org). Like Riesel Base 5 PRP I would like to see NPLB with team support.

2008-02-01, 06:53	#7
mdettweiler A Sunny Moo Aug 2007 USA (GMT-5) 1100001101001₂ Posts	I was just looking at the scoring page a moment ago, when I noticed that I was ranked right below CedricVonck for completed ranges--I had a slightly lower score (even though we each have done one range total so far), despite the fact that he had less candidates in his range than I did. And even though mine was a 2nd drive range (whereas his was a 1st drive range, and thus higher n), wouldn't it still be scored higher, because of the added bonus for searching non-top-5000 ranges?

2008-02-03, 04:39	#10
jasong "Jason Goatcher" Mar 2005 3×7×167 Posts	I can't help but note that after the Twin Prime Search primes fall off the list a few months from now, a person could find themselves in the position of actively crunching the project, maybe even increasing in rank, and yet getting lower and lower scores. Maybe we should archive a list, starting now, of the digit length of the prime at 5,000th place every first of the month. If someone gets discouraged, they have the option of making new comparisons based on a non-moving target, the value of the 5,000th placed prime in a given month. Is it possible to simply calculate a simple normalization value? (Not sure if normalization is the right term. )

2008-02-03, 05:54	#11
kar_bon Mar 2006 Germany B97₁₆ Posts	i'm calculating all scoring in Ex*el and the divisor of the formula is stored as constant in one cell i use. so i only have to edit '333333' into '350000' (or what ever) and the new calculations will be shown immedeatly! so, no problem. or do you meant a new formula at all?