Questions and stuff

[kar_bon]

he is changing to google-pages http://sites.google.com/site/sr2sieve/
but not all available yet there!

see his post here: http://www.mersenneforum.org/showpos...&postcount=523

[MyDogBuster]

he is changing to google-pages http://sites.google.com/site/sr2sieve/
but not all available yet there!

see his post here: http://www.mersenneforum.org/showpos...&postcount=523

Thanks Karsten

[Brucifer]

So what is the difference in the points formula between the regular database score computation and the prpnet score computation? I assume that the prpnet stuff is going to get added to the regular database point system?

[mdettweiler]

Originally Posted by Brucifer

So what is the difference in the points formula between the regular database score computation and the prpnet score computation? I assume that the prpnet stuff is going to get added to the regular database point system?

Yes, that's the plan. The only reason why it's not in there yet is because there's one last tweak Dave needs to do first, but once that's all set everything prior to that point will be added in retroactively.

The points will be calculated by the same formula; actually, at this point, the DB doesn't even know the difference between LLRnet and PRPnet results. We may at some point in the future have it do more advanced stuff with PRPnet results (which include a bit more information--such as the client name as well as username--than do LLRnet results), though none of that's even on the drawing board yet.

[Brucifer]

So why are they different to begin with???

[mdettweiler]

Originally Posted by Brucifer

So why are they different to begin with???

Er...I'm not sure what you mean. There's not really a "difference" per se; it's only that PRPnet records more information with each results than LLRnet does. However, the DB isn't set up to utilize that extra information at this time, since it was designed initially for LLRnet results. That's OK, as the extra information recorded by PRPnet is just extra icing on the cake, so to speak; what we end up doing is converting the results to LLRnet format with a script and then importing them to the DB.

[henryzz]

Originally Posted by mdettweiler

Er...I'm not sure what you mean. There's not really a "difference" per se; it's only that PRPnet records more information with each results than LLRnet does. However, the DB isn't set up to utilize that extra information at this time, since it was designed initially for LLRnet results. That's OK, as the extra information recorded by PRPnet is just extra icing on the cake, so to speak; what we end up doing is converting the results to LLRnet format with a script and then importing them to the DB.

i think he wants to know why we are changing to prpnet as llrnet has been working well for us:)

[Brucifer]

Actually :-) what I'm curious to is the number of points granted in the prpnet G3000 stats per pair returned, as it looks like 460+ per.... versus what is granted for a returned pair for the regular database points per returned pair.

As for prpnet, I like it personally. I ran it earlier when the initial beta round was going and it was stable for me. I've got it running on 12 cores now, granted they are mainly slow ones, but none the less they crunch out pairs. :-) I intend to add some fast cores to the prpnet and shift some off the port 7000 stuff and see if I can manage to find some big primes.

[TimSorbet]

Originally Posted by Brucifer

Actually :-) what I'm curious to is the number of points granted in the prpnet G3000 stats per pair returned, as it looks like 460+ per.... versus what is granted for a returned pair for the regular database points per returned pair.

As for prpnet, I like it personally. I ran it earlier when the initial beta round was going and it was stable for me. I've got it running on 12 cores now, granted they are mainly slow ones, but none the less they crunch out pairs. :-) I intend to add some fast cores to the prpnet and shift some off the port 7000 stuff and see if I can manage to find some big primes.

PRPnet:

How are user stats computed?
    The score is computed as such:
       (decimal length of the candidate / 10000) ^ 2

NPLB:

Participants initially ordered by Pair Score (n^2/160e9) descending.

I'm sure an approximate (should be quite close, but not precise, as NPLB only considers the n, allowing a small variation in decimal length resulting in the same score) conversion formula could be worked up. I'll see about trying to do that...

[TimSorbet]

I think the conversion factor is about 144.9905. That is, PRPnetScore ~= 144.9905 * NPLBScore, or PRPnetScore / 144.9905 ~= NPLBScore. Here's how I got that: (where c is the conversion factor)

(digits / 10000)^2=c*(n^2/160e9)
digits^2 / 1e8 = c*n^2/1600e8
digits^2 = c*n^2/1600
*
(n*log_10(2))^2 = c*n^2/1600
log_10(2)^2*n^2 = c*n^2/1600
log_10(2)^2*1600 = c
144.9905... ~= c

* Here I make the assumption that digits=n*log_10(2), which is decently close, but not perfect. Because of the differing algorithms, an exact comparison can not be made.

Edit: I noticed that I used log_2(10) where I should have used log_10(2), resulting in 17656.33 instead of 144.9905. It's fixed now, and is producing a number that matches Karsten's numbers below and makes more sense: 144.9905. This paragraph was from before I fixed it:
Hmm, this seems wrong...it would mean that your 1080 tests from the 5th drive, with a (PRPnet) score of 536349, are only worth about 30.377 (NPLB) points, but from looking at http://www.noprimeleftbehind.net/ind...r_pairs&page=2 that seems more like 6 decently sized candidates. Either I went very wrong somewhere (I don't see where), or one or both of the scoring formulas I assumed is not accurate. (I have verified that in the newest released PRPnet source, that's what it really does, so that really just leaves an error on my part or NPLB's n^2/160e9 formula being wrong) Can anyone see where I went wrong, and can one of the NPLB admins verify what the formula is?

[kar_bon]

the scores from G3000 are not correct as shown here:

G3000 for n~740k:

496 points / pair with PRPnet
3.42 points / pair with LLRnet