sieve stat question

**biwema** · 09-21-2003, 09:07 AM

hi,

I seem a bit unlucky though... Good thing the active window starts moving again soon. *rubbing hands*

I also noticed, that mklasson was sometimes quite unlucky. I always thought he will pass me next days because he did more sieving. The chance to find a factor within the window is now smaller as it is just 200k wide.
Mklasson, things are getting better for you when the window moves beyond 4.87M. You will pass me soon. I wish you all the best and good luck.

biwema

**mklasson** · 09-21-2003, 09:22 AM

Thanks!

**Joe O** · 09-21-2003, 02:50 PM

Originally posted by MikeH
I guess the only possibility is that a factor already exists for this candidate, but isn't present in the results or lowresults files.

Mike, When was the last time you got a refresh for the lowresults file?

**MikeH** · 09-21-2003, 04:03 PM

Mike, When was the last time you got a refresh for the lowresults file?

The one I'm currently using is dated 18th Jun. That file takes us from 1G to 3T, but I thought at that stage the whole range was complete. I guess it's possible that a factor was missed and then later picked up by P-1 or by someone looking very closely to try to find holes.

**Joe O** · 09-21-2003, 07:59 PM

Originally posted by MikeH
The one I'm currently using is dated 18th Jun. That file takes us from 1G to 3T, but I thought at that stage the whole range was complete. I guess it's possible that a factor was missed and then later picked up by P-1 or by someone looking very closely to try to find holes.

1598494103 | 33661*2^2975400+1

**Mystwalker** · 09-22-2003, 04:08 AM

ATM, there seems to be slight "disturbance" in the active window. The lower bounds dropped to a value below 4,600,000 again. I already witnessed this once before. I'm not sure where this originates, maybe a test that needs to be redone in the time frame the script checks the current bound?

One could change the script so the active window can't get smaller, or is there a reason against doing so?

But hey, thanks to this, I passed mklasson...

edit: Ok, active window changed to normal again and thus I'm at position 5 again...

**MikeH** · 09-22-2003, 01:30 PM

Originally posed by JoeO
1598494103 | 33661*2^2975400+1

I am confused. Why did you get this from?

**Joe O** · 09-22-2003, 05:36 PM

Originally posted by MikeH
I am confused. Who did you get this from

Mike, I got it from you last February. I have another 531 of yours, and 10 of mine that are not in your results

**Keroberts1** · 09-25-2003, 12:47 AM

any updates on this?

Originally posted by MikeH
I'm hoping that maybe the PRP=1 is not true, and is somehow a minor bug in the assigment code. But my worst fear is that this there is a much more serious bug in the assignment code, and this test really was reassigned even though a factor was present at the time of reassignment.

**MikeH** · 09-28-2003, 05:04 PM

Mike, I got it from you last February. I have another 531 of yours, and 10 of mine that are not in your results

Many thanks Joe. Brain fade on my part.

All of those are now incorporated (which is why the stats are a little distorted). Now I just need to take away that one that scored almost 2000 for me, and all will be square.

**Joe O** · 09-30-2003, 07:50 PM

Originally posted by MikeH
Many thanks Joe. Brain fade on my part.

Mike, You're welcome!

ps what's with the Gap Analyses?

None of the ranges:
1.5T<p<60T 0T<p<60T

60T<p<80T 60T<p<80T

...

220T<p<240T 220T<p<240T

240T<p<2000T 240T<p<2000T

correspond to the pages they point to:
Gaps analysis 1.500T < p < 10.000T (0.3M < n < 3.0M)
Gaps analysis 10.000T < p < 20.000T (0.3M < n < 3.0M)
...
Gaps analysis 90.000T < p < 100.000T (0.3M < n < 3.0M)
Gaps analysis 100.000T < p < 2000.000T (0.3M < n < 3.0M)

nor have they been updated recently:

"Last Update: Sun 28-Sep-2003 16:24 (GMT+1)"

**ceselb** · 10-01-2003, 03:28 AM

The not updating part might be because of me.

I did some reformatting of the coordination pages, that might have broken Mikes scripts.

**Keroberts1** · 10-01-2003, 06:18 AM

I've been wondering about this for a while and I would like to know who could answer this. If a factor is found for a certain N value that is currently being tested but the test is aborted will it be reassigned or not. I know this was uncertain before i was just wondering if anyone has checked on this because I have had several factors that have landed in that range and stayed on my stats page for a while only to dissapear once i was convinced that the test for it must have been aborted.

**MikeH** · 10-01-2003, 07:48 AM

ps what's with the Gap Analyses?

Sorry, looks like I've broken it. As you can tell I changed all the ranges at the weekend, and when I tested it locally it worked. I guess the pages aren't being sent up to the server.

I'll take a look at it this evening.

P.S. ceselb, it's nothing you did, honest.

**ceselb** · 10-01-2003, 08:41 AM

Originally posted by MikeH
P.S. ceselb, it's nothing you did, honest.

Ok, nice. I was kind of relying on your scripts being flexible enough to handle the change.

**Joe O** · 10-01-2003, 10:03 AM

Originally posted by MikeH
I guess the pages aren't being sent up to the server.

Or you forgot to change the underlying pointers: eg for "200T<p<220T"
"http://www.aooq73.dsl.pipex.com/gaps/Gaps_n3_20_p08u.htm" to "http://www.aooq73.dsl.pipex.com/gaps/Gaps_n3_20_p20u.htm" or whatever your new naming scheme is.

**MikeH** · 10-01-2003, 12:40 PM

OK, gap analysis is working again.

(I had commented out the line in the script that uploads the pages while I did some quick tests

)

**Keroberts1** · 10-02-2003, 07:49 PM

earlier on my stats showed taht the range i had to finish from 87200-89970 had around 80 missing factors however at 87246-89970 it says about 600 factors. Is that right. I have noticed that factors were very common for a while, I got a whole bunch probably like 14 in a course of about 30 when i regularly see gaps of like 8-14. Is it possible taht factor density could follow trends other than strictly loosing density. Could some ranges inherently have higher factor density than others. Maybe we could find a formula to help us decode this possible pattern. I believe that the stats change was probably due to the density at that specific range If I'm wrong my bad.

I don't really know anything. In any case the density of that range is surprising and I gotta wonder the odds of that happening again. The range did nothing for my score by the way.

**mklasson** · 10-03-2003, 06:57 AM

MikeH,
saaaay, what's up with -17K under "Other"?

Code:

Score breakdown
                               Total   New factors (    %)  Score change (    %)         Other (    %)  
Daily change last  1 day :    -1903.39         24.79 ( -1.3)      15528.65 (-815.8)     -17456.83 (917.1)

**MikeH** · 10-03-2003, 09:12 AM

saaaay, what's up with -17K under "Other"?

It would appear that all the hugh factors have disapeared from the results.txt file. (Louie/Dave what's going on).

As a result, the following have been dropped from the scores. Don't worry though, the finds are still recorded in my DB, so when they appear in the results file again, normality will be restored.

1335945347757694154470789 28433 2473 2876 150
1420021239577240035601 10223 4187705 3109 203
148603027109277394117 67607 4090091 769 0
179928189648359393301449201 55459 1054 3031 81
192319539630928827473 5359 10150 2876 150
249138319295011519189933003 22699 1414 2876 150
25756370825971135421 55459 316750 2876 150
273148210774616633431 5359 4151886 365 60
27528683020268847973 21181 4829108 1608 0
29132891782118156447 33661 323016 2876 150
3553460704416618023888371 55459 1030 2876 150
38681107049404634813 33661 1584 2876 150
39480881218898286589 10223 8105 2876 150
42498445799169841757640079 5359 2230 2876 150
552467210540360947721 4847 304983 2537 0
771852543150955945417 10223 3821 2876 150
94846070521408174091 10223 4348601 584 61

**MikeH** · 10-03-2003, 12:14 PM

It would appear that all the hugh factors have disapeared from the results.txt file.

Now it's fixed. Did anyone change anything?

**MikeH** · 10-03-2003, 12:33 PM

earlier on my stats showed that the range i had to finish from 87200-89970 had around 80 missing factors however at 87246-89970 it says about 600 factors. Is that right.

The estimate for the number of factors in a given range is based on the density of factors in the preceding ranges that it thinks have no gaps. Since the areas preceding your range are very disjointed in their completeness, the estimates will have been a bit questionable. As more factors have come in (including your own submissions at the beginning of this range), the estimates will have got better. The current estimate looks about right - 570 factors for a 2705G range - 0.2 factors per G.

**MJX** · 10-07-2003, 07:14 AM

I think that the stats, as they currently are, aren't very attractive for newcomers, even with high cpu power (and that isn't my case).
Imagine someone with 4 or 5 athlons boxes, sieving at 5*400=2000 kp/s and starting at 100T. The discovery rate should be around 40 factors every 250G, meaning 1 factor in the active windows (and immediately scoring) every 250G... So this person will score something that 40000 a week (5 "big" ones at 7000 and a few others)...not bad but...
It will be hard for him to reach the Top10 and, morehower, he won't benefit the huge "stock option" power cumulated by those who were sieving at 10T, with a discovery rate ten times higher, and that have dozens of factors with n below 5000000 waiting to see their score explode soon..
He will NEVER be able to progress and reach the top and, if he is stat addict, will certainly not join the sieve yet...
This state is inherent to the rarefaction of factors discovered, so the sieving effort is exponential and not lienar (as in DF or Seti or many others)...
So I suggest to segment the stats, for example by year. One should close the stats at the end of 2003, letting the scores growing so that the former work will benefit to the "2003 challengers", AND open a 2004 stats page from scratch, by users and team, so that a new challenge may begin at 110-120T, with the ability to be a newbee, or a new team, and to reach a good place in stats....
This is still done in projects like ECMNET because hitting factoring records with ECM means an exponential difficulty too...

**Mystwalker** · 10-07-2003, 07:33 AM

You have to consider that factors found earlier have a relative small factor value. This means the bias is a lot lower than that of current (and future) findings. This negates the disadvantage of a worsening factor density, as it proportionally affects the score of a factor.

So the score for a factor is basically independant from the effort needed to find it.

**MJX** · 10-07-2003, 01:02 PM

I do agree, but when the top score is 630000, and will probably increase a lot with the active window, I think that even with 7300 pts scoring factors the challenge is a bit too hard for new members...(and, basically, it is funnier to find many "small" factors and see their growth than to get one big every week and see the others score climbing...). So I persist in my suggestion of a renewal of the stats for 2004...
(a project as the XYYX factorization has lost most of this members last year because it was becoming too difficult and got a big boost when an extension was releasing, offering many news easy small factors to find...most people doing DC for fun and not for the state of the art!)
Best wishes.

**Frodo42** · 10-07-2003, 01:22 PM

With P-1 factoring it's actually not that hard to get into the top-10. I've almost only done P-1 for about a month with a 2.4 GHz P4 and already I'm in the top 10. I'm not sure that I'm going to stay there, as I almost don't have any "stock" and the window has now started moving, but I have not fallen as fast as expected since those factors that already scored keep on getting more points as the bias climps (which I don't really contribute to ...).

I think the important thing about the stats is that it should reflect the positive effect the work you do have on the SoB project. How to weigh the positive effect you made in the past with the one you make now is a little difficult.
The system with moving windows is very good thing as it makes the scores reflect the actual postive effect.

**MikeH** · 10-07-2003, 02:51 PM

I have to say I'd tend to agree with Mystwalker and Frodo. Those stock options are becoming worth very little at an alarming rate. With new factors now scoring over 7000, I think that with two fast P4s you could P-1 factor your way to top five or top maybe even top three in the 14 day average.

I'd have to agree that the "all time" scores are now becoming a bit stale, but the 14 day average should really be viewed as the place to be, and shouldn't be too big a challenge for any new comer.

Having said all that, I have no objections to having (say) a quarterly challenge, where only factors submitted in that period would score towards that challenge.

The one thing I don't want to encourage is people holding back factors. If everyone knows that the next challenge will start at the beginning of 2004, then I'm sure that some people will hold back factors until then. Although in itself this isn't a problem, ultimately it will lead to factors being forgotten or lost, and that's the last thing we all want.

**Joe O** · 10-07-2003, 03:18 PM

Originally posted by MikeH
If everyone knows that the next challenge will start at the beginning of 2004, then I'm sure that some people will hold back factors until then. Although in itself this isn't a problem, ultimately it will lead to factors being forgotten or lost, and that's the last thing we all want.

The next challenge can start when we start 20M to 50M.

**MJX** · 10-07-2003, 03:54 PM

If everyone knows that the next challenge will start at the beginning of 2004, then I'm sure that some people will hold back factors until then.

ohohoh...I didn't thought people may be so pervert...

ok ok for the global stats and challenging for the last 14 days..

But the idea of quarterly challenges sounds great to me. Imagine " the christmas ulltimate week" where everybody is challenging and try to break the records (for example by stealing your boss cpu cycles while he is on hollidays, installing an hidden client on his desktop...) what a fun!!!

best wishes.

ps : stealing cycles is bad and it was a joke, and i know one shouldn't use his boss cpu cycles without permission... etc...etc...

**Nuri** · 10-14-2003, 02:28 PM

Mystwalker, congratulations for the first place in last 14 days stats.

I'm sure you'll soon get the second place from me in the overall stats.

**MikeH** · 10-14-2003, 03:15 PM

Mystwalker, I too would like to congratulate you.

You have some serious sieving power right now (and P-1 factoring too).

**Mystwalker** · 10-14-2003, 03:46 PM

Originally posted by Nuri
I'm sure you'll soon get the second place from me in the overall stats.

I hope so.

But I doubt I'll stay at #1 for long - semester starts now, so I won't have that much computing power at hand anymore.

For a short time, I had 17-18 machines sieving plus 1 P4 factoring. The first ones are (except one 1 GHz) only 800 or even approx. 450 MHz, but they are running 24/7 and the sheer bulk converted them into a big mean crunching machine.

And, of course, I have to congratulate mklasson for his great sieving client, as it as fast as lightning and provides several features to ease administration a lot as well.

Ok, so now, you all have the goal to beat me again.

I think we all contribute a lot to this project and can be proud of it...

**MJX** · 10-16-2003, 07:52 AM

hello!

in the stats page, user stats, a range completed more than a week ago is still noted incomplete in the 300K-3M range and the new reserved one doesn't appear...So, factors found in this range are marked as non reserved ones... The forum's coordination pages are up to date so I can't understand what is the problem...
Is the update an automated task in the stats page or is it done by hand? May I help by submitting to MikeH my completed and reserved ranges???

Thanks.

**MikeH** · 10-16-2003, 01:31 PM

Is the update an automated task in the stats page or is it done by hand?

Sorry to say that this is a manual task. It's about the only manual task, and I do just block cut and paste from the forum pages, but still it is manual.

I tried automating it a while back, but decided it was more trouble than it was worth. Things seem to have settled down a bit since then, so maybe I'll give it another try.

In any case, I try to do an update at least once a week (usually much more often), but I only take the main post, not the recent additions at the bottom.

I've just updated, so in the update in 2.5 hours it should be OK.

Also remember that as long as you're logged in when you submit factors, then it isn't too important that your reservation isn't shown. If you're not logged in it can be a big problem, because a score won't be awarded until it can identify the user - if a factor is close to the PRP wave, then the wave may have crashed over before I've had chance to do an update.

**ceselb** · 10-17-2003, 07:25 AM

Nice, than maybe you can look into this:
A range is marked incomplete for me. Could you check 16250-16270 for me?

Some ranges that are abandoned in p-1 are marked as incomplete also (atleast I think they are).
Maybe move them to complete, but keep the [abandoned] bit.

No hurry.

**Keroberts1** · 10-26-2003, 04:48 AM

Has there been any updates on the factors found after tests have been assigned but abandoned. I just foundanother factor today about 6 hours after it got assigned out. If I'd checked the system earlier i would haveprobably gotten it in in time. Ohh well... Jus wondering if the bugs had been worked out with that. Maybe we could add a feature to keep a list of the tests being performed that have factors and the name of the user doing the prp. Then the staff at SoB could attempt to manually remove the canidate to prevent it fom being reassigned? This is all just hopeful thinking. I really don't know if this is possible. Well I'd at least love to hear if anything is known wether or not a factor in the testing prp range is worth anything at all.

Side thought...

man I haven't found any useful factors in weeks i really need this.
Frodos overtaken me.

**Frodo42** · 10-30-2003, 03:15 AM

For some reason my last found factor
11431293770421995731 | 27653*2^5103393+1
does not seem to be anywhere to be found in the stats.
It's more than 24 hours since I submitted it and Keroberts1 overtook me in the meantime

**MikeH** · 10-30-2003, 03:45 AM

does not seem to be anywhere to be found in the stats.

That's because it's not in the results file. But why it's not in the results file I don't know. I tried resubmitting the factor, and it wasn't added to the DB, so it is already known.

Dave or Louie will need to take a look at this one.

Edit: Just submitted it using the largesieve page, and it claimed to have added it to the DB, so we'll see if it turns up in the update in 6 hours from now (in my name). I can sort out the owner afterwards

Edit (again): ...and before anyone asks why their scores have gone down, the largeseive factors disapeared from the results file in the 09:00 GMT update.

**Keroberts1** · 10-31-2003, 04:27 AM

Most recently (significant) changed scores
p (T) k n Score Factor found Score changed Score was Score could be Reqd bias
88.150 67607 4940451 6723.629 Sun 26-Oct-2003 Fri 31-Oct-2003 0.881
88.105 10223 5180765 7389.934 Sun 26-Oct-2003 Thu 30-Oct-2003 0.881

This was in the next to pass Upperbound but i gues the test as stopped and i got scored for it?

**MikeH** · 10-31-2003, 10:36 AM

This was in the next to pass Upperbound but i gues the test as stopped and i got scored for it?

More likely that the 'next n' jumped arround and ended up lower than this, so you'd score. The test is probably still on-going, but hey, you got lucky.

Thread: sieve stat question

Thread Tools

Rate This Thread

Display

Posting Permissions