Any possibility or merging stats PRP, Sieve, P-1

[vjs]

I was wondering if there is any possibility of merging the stats for the main prp effort with those points accumulated by sieve or factoring???

Main reason ... some people do a no-net dc, but sieving or prp isn't that attractive to the stats people.

[kugano]

Yes. This is one of the main goals of version 3.

[vjs]

Will the stats be reset to zero or how will existing credit be implemented.

Major reason I'm going to be asking some of my group ARS to come to sob sieve or factor

[kugano]

Stats will be fully integrated with a unified point system. The point units will completely replace cEMs and will be the same whether you're doing PRP tests, factoring runs or sieve ranges. All work and power measurements will be made with these new units (along with the usual conversions to flops, P90-years, all that fun stuff), with by-work-type breakdowns showing how much of the total work / power is accounted for by PRP tests, factoring, etc.

Stats will not be reset to zero. However, they will be adjusted to accomodate the new system of measurement. For example, old cEM values for PRP tests calculated using O(n^3) will be retroactively altered by switching to O(n^2 log n).

The consequences of this are that more credit will be given for "old" work than is currently done. So suppose two teams, A and B, are exactly even in score. But suppose Team A did most of its production a year ago, and then petered out, while Team B didn't even exist a year ago, and has only recently caught up to A. The stats adjustment will cause Team A to shoot ahead of Team B.

This may sound unfair... and it's definitely unfortunate. But I want to stress that the current statistics are inaccurate, and the adjustments are made to improve accuracy. So Team B in the example might be upset, but they need to realize that they really were behind Team A all along – it's just that the current stats are so innacurate that they misleadingly indicated the two teams were even.

This retroactive adjustment is still a ways off though, so don't panic just yet ;-)

[Troodon]

How the factoring and the sieving stats will be converted?
Will you count in any way the sieving/factoring time/power in the new stats system? Currently MikeH's stats give points only if you find factors, which is not very bad for sieving, but for factoring is a bit unfair (someone could factor 10 k/n pairs and get two factors, which means points, but someone else could factor 100 k/n pairs and don't get any point if he/she doesn't find a factor).

[kugano]

You'll get flat-rate credit, regardless of whether or not you find any factors. We'll do our best to arrange it so that a prp test that takes X seconds will give you exactly the same amount of credit as sieving for X seconds on the same machine. It won't be exact... one may give you very slightly more points than the other, but we'll do the best we can to even it up.

I'm not sure what we'll do, if anything, about retroactive credit for sieve ranges. We might be able to coerce the range histories that are posted on the forums into something we can generate stats from and store in the database... but I make no promises on that.

[Keroberts1]

perhaps it would be a good idea to rate scoreing value o nthe value of the effort to the project. Personally i like the sieving stats the way they are now. And the moving active range is alot of fun. perhaps we could just have the sieving stats remain the way they are and have a conversion ratio that will be used. That way we could still have some strictly sieving stats.

[kugano]

You still will have some "sieving only" stats. Just because they're integrated doesn't mean they can't be broken down into specific types of stats for PRP tests, specific stats for factoring, sieving, etc.

[Lagardo]

Originally posted by kugano
[B] We'll do our best to arrange it so that a prp test that takes X seconds will give you exactly the same amount of credit as sieving for X seconds on the same machine. It won't be exact... one may give you very slightly more points than the other, but we'll do the best we can to even it up.

In that case let me urge you to err slightly on the side of giving more credit for PRPin than for sieving -- otherwise large crowds will start sieving just to crank up their scores ever-so-slightly faster. And if I remember the last resource-allocation thread right, the current ratio of people who PRP vs. sieve is just about right...

[kugano]

Sounds good to me.

[dmbrubac]

And on that note, perhaps the credit should be variable. In other words, if it's 'just right' between PRP, sieving and factoring, the credit is equal. If there are too many sievers, credit drops, making it less attractive. Likewise, if there are too few factorers, credit increases (as is the case right now! If you can spare a P4 we could use the help. There are only 3 active factoring people right now). The opposite would be true in both cases, of course. PRP should not be adjusted, since it is really the main reason for doing all this.
I know it adds some complexity, but it could provide excellent self levelling capabilities.

[kugano]

I don't like that idea at all.

The "self-levelling" is exactly why I created the priority fields in the queueing system. The balances between different types of work should be controlled server-side by adjusting the priority values so that the server always hands out the best tests. And that's much simpler than mucking around with work scores and waiting for the user pool to catch on and switch their preferences around.