mersenneforum.org > Data And now for some TF results...
 Register FAQ Search Today's Posts Mark Forums Read

 2019-02-23, 20:49 #23 Mark Rose     "/X\(‘-‘)/X\" Jan 2013 Ͳօɾօղէօ 22·5·139 Posts Couple thoughts. It's pretty easy to guard against most throwaway email domains by using https://github.com/ivolo/disposable-email-domains We could move to giving TF credit only when a factor is found. We have enough data to figure out the probabilities of finding a factor, so we can estimate the work done to find the factor. Disadvantages are not seeing when someone submits a bunch of bogus lines, and it could be discouraging to new users... though most people doing TF have been doing LL for a while...
2019-02-23, 20:52   #24
SELROC

2×5×7×112 Posts

Quote:
 Originally Posted by Mark Rose Couple thoughts. It's pretty easy to guard against most throwaway email domains by using https://github.com/ivolo/disposable-email-domains We could move to giving TF credit only when a factor is found. We have enough data to figure out the probabilities of finding a factor, so we can estimate the work done to find the factor. Disadvantages are not seeing when someone submits a bunch of bogus lines, and it could be discouraging to new users... though most people doing TF have been doing LL for a while...

That would make my trial-factoring today totally useless and unfructuous as no factor found !!!

2019-02-23, 20:56   #25
retina
Undefined

"The unspeakable one"
Jun 2006
My evil lair

23·3·13·17 Posts

Quote:
 Originally Posted by Mark Rose It's pretty easy to guard against most throwaway email domains by using https://github.com/ivolo/disposable-email-domains
I would resist against using such lists since it excludes people that have trouble getting a "real" email address that requires registering a phone number. It's troublesome for privacy and not everyone has (or wants) a cell phone.

2019-02-23, 20:59   #26
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

54118 Posts

Quote:
 Originally Posted by Mark Rose We could move to giving TF credit only when a factor is found.
I'd be in favour of this. It both obviates the problem of people submitting false negative results to boost the credit score of their fake account (a motivation that eludes me, but apparently exists), and it more accurately reflects reality: factors are the real prize in TF in that they positively disprove primality; the fact that they're trivially verifiable is a bonus. No-factor results are not worth anything in and of themselves, just a stepping stone towards a collective factor-finding effort.
Quote:
 Originally Posted by SELROC That would make my trial-factoring today totally useless and unfructuous as no factor found !!!
Not at all. Over time you will have some days where you find factors and some where you don't, but when you do find a factor it would be worth the estimated effort to find it (the time to run the actual TF that found the factor, plus the other ~70 TF runs that found no factor). No-factor result value drops to zero, Factor result value jumps by about 70x.

2019-02-23, 21:13   #27
Mark Rose

"/X\(‘-‘)/X\"
Jan 2013
Ͳօɾօղէօ

22×5×139 Posts

Quote:
 Originally Posted by retina I would resist against using such lists since it excludes people that have trouble getting a "real" email address that requires registering a phone number. It's troublesome for privacy and not everyone has (or wants) a cell phone.
I guess the major free providers have started requiring numbers, haven't they?

2019-02-23, 21:15   #28
SELROC

26A016 Posts

Quote:
 Originally Posted by James Heinrich I'd be in favour of this. It both obviates the problem of people submitting false negative results to boost the credit score of their fake account (a motivation that eludes me, but apparently exists), and it more accurately reflects reality: factors are the real prize in TF in that they positively disprove primality; the fact that they're trivially verifiable is a bonus. No-factor results are not worth anything in and of themselves, just a stepping stone towards a collective factor-finding effort. Not at all. Over time you will have some days where you find factors and some where you don't, but when you do find a factor it would be worth the estimated effort to find it (the time to run the actual TF that found the factor, plus the other ~70 TF runs that found no factor). No-factor result value drops to zero, Factor result value jumps by about 70x.

but in some way by dropping the value to 0 you effectively hide GHz/day information.

I find a better way to deal with this is to require email verification !

2019-02-23, 21:23   #29
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

52×113 Posts

Quote:
 Originally Posted by SELROC but in some way by dropping the value to 0 you effectively hide GHz/day information.
For no-factor results, yes. Overall for any TF contributor the running average throughput will remain unchanged.

Quote:
 Originally Posted by SELROC I find a better way to deal with this is to require email verification !
Certainly a worthy first step, but I can guarantee that anyone who wants to be a nuisance will not be deterred by such simple things.

2019-02-24, 00:16   #30
R. Gerbicz

"Robert Gerbicz"
Oct 2005
Hungary

1,327 Posts
Foolproof on TF (5 sigma)

Quote:
 Originally Posted by LaurV Email verification should work (well.. at least, as you said, slow down some badass kids).
Quote:
 Originally Posted by Madpoo If it were asymmetrical so the server could quickly verify, that'd be okay. One glitch is that there are users who genuinely submit a LOT of results daily (think TJAOI or even the Gpu72 bot) so it might place an undue burden on them.
Obviously the hard part is that for no factor (what is the common) there is nothing to check in the usual codes.

In the past I've thought that it is just impossible,
don't have an exact idea what would be the slowdown on gpu (I'd say that less than 1%),
but the check on the server would be terrible fast and close to impossible to fool it:

Let x mod N the unique mod N residue in [0,N).

and H=hash(p,q,t)=(2^p mod q) mod (2^t)

we are expecting to see H=1 for cnt/p/2^(t+1) different q primes (q=1,7 mod 8) if there are cnt primes in [2^b,2^(b+1)). Save these q numbers if q is prime ( or pseudoprime ) and send them to the server at the end of the computation.

The number of hits follows a binomial distribution with:
Binom(N,r)=Binom(cnt/2/p,1/2^t)

Code:
where:
N=cnt/(2*p)
r=1/2^t
ev=N*r (the expected value)
sigma=sqrt(N*r*(1-r))
here mark the submission as suspect if you'd see less than ev-5*sigma different(!) q values from the [2^b,2^(b+1)) interval, where q mod 8={1,7} and q is prime (or pseudoprime, that would change almost nothing). Ofcourse accept the submission for p, if it has found at least one q>1 divisor of mp (because it is possible that in this case the search used an early abort strategy at success, what is common).

choose such t for that say 1000<ev<2000 (there'll be a unique t if N>1000, and if the range is not very small).

ps. For large bounds we have exact prime counts for [2^b,2^(b+1)), but you can also use estimations. Don't count composite q values, because that mess up these things, depends on your sieving bound etc.
Why would be hard to fool it? Because for actual q prime divisors of mp we have also H=1, if you could find these q much faster, than the whole tf would be also much faster.

Last fiddled with by R. Gerbicz on 2019-02-24 at 00:19 Reason: typo

2019-02-24, 01:03   #31
Serpentine Vermin Jar

Jul 2014

3·5·7·31 Posts

Quote:
 Originally Posted by kriesel None of the gpu apps have interfaced the Primenet API or security code. I think there are some structural issues with the cpu-oriented primenet API that relate to that. I suppose one could give the primenet api fake properties for a gpu as a strange sort of single-core cpu to try to shoehorn it in.
Pretty much, yeah. When I was noodling around at it, I created a new computer using the API (constructing my query string manually, with the advantage that I can bypass the security features for myself).

I just specified the cpu type as "Nvidia,RTX2080" just to have something in there (a true, standard format should be adopted so any devs are agreeing to a common thing like "Manufacturer,Model,whatever")

Some other things like L1/L2 cache sizes, # of cores, I just set to 0 for cache, 1 for # of cores. The exact values there don't matter so much although it could come into play during assignments when trying to see if an assignment could feasibly complete in X amount of time.

Other things like the "feature set"... not sure what would be agreeable for GPUs there since I'm not a GPU guy.

Point being there, I was able to shove in a new computer account that way.

2019-02-24, 01:14   #32
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

17·251 Posts

Quote:
 Originally Posted by James Heinrich Over time you will have some days where you find factors and some where you don't, but when you do find a factor it would be worth the estimated effort to find it (the time to run the actual TF that found the factor, plus the other ~70 TF runs that found no factor). No-factor result value drops to zero, Factor result value jumps by about 70x.
Averages are good in theory over the long run.. however I see a big problem here. The actual stats are at home but I did a complete TF summary at year end (everyone).

There were hundreds or maybe over a thousand TF users who did not report a single factor; some with several hundred attempts. They would get a big ZERO thanks for nothing and probably never return. Then there were several dozen users with success rates well well over 1% (not only TJAOI at 99%) but a few who it seems don't bother to report NF. These would get relatively very large credit.

2019-02-24, 01:15   #33
Serpentine Vermin Jar

Jul 2014

CB716 Posts

Quote:
 Originally Posted by SELROC but in some way by dropping the value to 0 you effectively hide GHz/day information. ...
Not necessarily... in case you missed it, what they're suggesting is that "no factor found" has zero credit given, but when you do find a credit, it's a LOT more than it is right now.

If you're doing TF and not finding any factors, there's something wrong there (or in the case of these fraudsters, they're just making it up as they go).

I'm not convinced whoever is doing this is just doing it for the credit, since it's essentially meaningless to anyone outside the Mersenne prime finder community, and it's a prestige thing really. Not sure what someone things they'll gain by doing this besides just being a <insert whatever bad word here>

 Similar Threads Thread Thread Starter Forum Replies Last Post ET_ Operazione Doppi Mersennes 604 2020-03-26 15:17 lycorn PrimeNet 22 2017-10-02 02:40 danaj Prime Gap Searches 0 2017-08-14 18:35 Unregistered Information & Answers 3 2010-07-26 00:49 Mike PrimeNet 11 2004-05-23 12:55

All times are UTC. The time now is 00:19.

Thu Apr 2 00:19:29 UTC 2020 up 7 days, 21:52, 3 users, load averages: 2.03, 1.67, 1.40