Page 4 of 5 FirstFirst 12345 LastLast
Results 121 to 160 of 172

Thread: sieve stat question

  1. #121
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    Mike:

    I've found a strange occurence on my personal scores page:

    Score change is said to be 13549.09 last day, but there was only one change: 1.347 --> 13555.303
    Shouldn't the score change be 13553.956 (other score changes would even increase this gap - unless I had negative score changes, of course...).

    Do you have more insight?

  2. #122
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    Do you have more insight?
    When is a day not a day? That is the question.

    Having just taken a look at this, it would appear that the problem occurs because one day is actually taken as the time between this update and the one that's about 24 hours ago. Because the clock doesn't start until the various files that are required for the update have been downloaded, there can be a variation of many minites. In the data currently displayed, the gap between updates was 24.04 hours.

    You'd expect the score change to be 13555.303 - 1.347 = 13553.956

    but adjusting for a day: 13553.956 * (24/24.04) = 13531.403 (which is what's displayed)

    This adjusting for a day feature was designed for when the score update fails totally for a number of runs. When that happens it makes slightly more sense because the two comparison points will be much more than 24 hours appart.

    Hope that helps.

  3. #123
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    Originally posted by MikeH
    Hope that helps.
    Definitely! Thanks for the clarification!

  4. #124
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    Mike,

    It seems to me that, since we started to have secret and supersecret doing DC work on two different ranges, I guess we should have a slight change in our sieve scoring system.

    In short, we need a third "active window" on the 2m+ range to reflect secret related effects.

    To complicate things a bit more, we'll also need a second completed window for 2000000<n<2m+


    PS: Also, since we've finished all original secret tests, we will not have a case for
    "p > 40T, in 'DC active' window, 0 PRP tests performed," anymore.

  5. #125
    Hater of webboards
    Join Date
    Feb 2003
    Location
    København, Denmark
    Posts
    205

    Thumbs down What happened to the scores?

    Something weird has happened to the scores, almost everybody lost lots of points. Mike is still in first place, but only with around 125000 points, and among the really weird stuff Kroberts5 (Keroberts1) only has 13433 points but under largest scores he has a factor giving 19808?

    BTW: Mike you seem to have missed that I actually have completed the 153300-153500 sieving range. (it's in the archived ranges thread now)

  6. #126
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    Scores are better since the last update, but "Current 90% sieve point: -0.001T" looks slightly incorrect.

  7. #127
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    Something weird has happened to the scores, almost everybody lost lots of points. Mike is still in first place, but only with around 125000 points, and among the really weird stuff Kroberts5 (Keroberts1) only has 13433 points but under largest scores he has a factor giving 19808?
    I've seen this before, looks like the results file it downloaded for the 09:00 update was almost empty. Looks like the 15:00 one was OK, the 90% point etc. will be off, because that comes from the previous run. Should all sort itself out at 21:00.

    Hmm....I really must add some protection against this happening (no update at all would be better than a crazy update).

    BTW: Mike you seem to have missed that I actually have completed the 153300-153500 sieving range. (it's in the archived ranges thread now)
    Sorry, I always miss the archived ones.

    BTW, I'm out of town from tomorrow for a week (Iceland in February ) , so I won't be able to updates my local copies of the sieve and factor reservations. But that won't be a problem as log as everyone remembers to log in before submitting factors

  8. #128
    Unholy Undead Death's Avatar
    Join Date
    Sep 2003
    Location
    Kyiv, Ukraine
    Posts
    907
    Blog Entries
    1
    BTW, I'm out of town from tomorrow for a week (Iceland in February ) , so I won't be able to updates my local copies of the sieve and factor reservations. But that won't be a problem as log as everyone remembers to log in before submitting factors [/B]
    Don't worry daddy, we will....
    wbr, Me. Dead J. Dona \


  9. #129
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    OK, scores are now sorted.

    Hmm....I really must add some protection against this happening (no update at all would be better than a crazy update).
    ...and I've done this too.

  10. #130
    Hater of webboards
    Join Date
    Feb 2003
    Location
    København, Denmark
    Posts
    205

    Exclamation A suggestion for Mike

    In the table listing a users largest factors T/P/E/Z/Y-suffixes are used to display the factors. It would be nice if these were used in the other tables (such as largest score, factors next to enter ...) too.

  11. #131
    178.600 67607 13126211 1.786 Fri 30-Jan-2004
    178.560 27653 2628537 1.786 Fri 30-Jan-2004
    178.516 55459 1843894 1.785 Fri 30-Jan-2004

    RATIO: 178491.08G - 178600.28G, size: 109.21G, est fact: 7 (178491077368091-178600284728843) R: 1.159, S: 0.672
    #*# ( 100.28G) : 178500-179000 cooper76 [complete]

    is something wrong with this ?

  12. #132
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    Both of the factors

    178.560 27653 2628537 1.786 Fri 30-Jan-2004
    178.516 55459 1843894 1.785 Fri 30-Jan-2004


    have n<3m. So, it's normal that they don't show up in the http://www.aooq73.dsl.pipex.com/gaps...n3_20_p07u.htm page, but show up in the http://www.aooq73.dsl.pipex.com/gaps...n3_20_ps0u.htm page.

  13. #133
    Senior Member Frodo42's Avatar
    Join Date
    Nov 2002
    Location
    Jutland, Denmark
    Posts
    299
    The sieveing scores seem to have taken an unexpected turn ... the current active window is listed to be -1 < n < 4830000

    Last time there was a problem with the active windows it was due to a prime, I hope that is the case again.

  14. #134
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    The project data from SoB looks ok, so I'd rather guess it's either an error with the results.txt or the scoring script.

  15. #135
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    In the table listing a users largest factors T/P/E/Z/Y-suffixes are used to display the factors. It would be nice if these were used in the other tables (such as largest score, factors next to enter ...) too.
    Done.
    The sieveing scores seem to have taken an unexpected turn
    ...and that's fixed too.

  16. #136
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    Damnit! I thought I almost got Nuri (score-wise), but now I'm 100,000 points behind again.

  17. #137
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    Originally posted by Mystwalker
    Damnit! I thought I almost got Nuri (score-wise), but now I'm 100,000 points behind again.
    It's for sure that you will one day. All I can hope is it will be far into the future.

  18. #138
    is there an active range for the sieve stats taht will allow tests for secret to score for the tests they have saved? Around 2 million? I don't see any i nthe stats page.

  19. #139
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    is there an active range for the sieve stats taht will allow tests for secret to score for the tests they have saved? Around 2 million? I don't see any i nthe stats page.
    'secret' is now a secondary double check window at n=2M, so really there are no more true 'secret' tests to save (in the old sense).

    I was going to add a second DC window to the stats, but there seems to have been so little activity in 'secret' since it ran out of first time tests that there seems to be little point. Looking just now, three tests were completed yesterday - that's far more than when I looked a couple of weeks ago.

    I'll have a think - the only real problem right now is determining where the window should be. I'll keep an eye on http://www.seventeenorbust.com/secret/ to see if this is behaves

  20. #140
    Hater of webboards
    Join Date
    Feb 2003
    Location
    København, Denmark
    Posts
    205
    I can see that we now have a third window (currently at 2002699 < n < 2012699), could we have a list of the factors next to enter this window?

  21. #141
    These scores don't make sense shouldn't the higher n value higher p value score higher


    197.675T 28433 2011777 500.025 Wed 14-Jan-2004 1.977
    86.435T 27653 2005665 651.942 Tue 14-Oct-2003 0.864


    also I've noticed some factors on one name that i thought was combined with my other name i use now. They didn't show up in the sats for next to enter main active window on the score page where i rank 6th i believe. Could someone please check to see if the factors saved under keroberts1 (id=3172) are properly credited ot the other name i use more often when submitting factors Kroberts5 (id=3423).
    Last edited by Keroberts1; 02-24-2004 at 01:56 AM.

  22. #142
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    197.675T 28433 2011777 500.025 Wed 14-Jan-2004 1.977 has 2 PRP tests performed.

    Thus; score = (n/1M ^ 2) * 125 * 0.2 * bias

    =((2011777/1000000)^2)*125*(197.675T/40T)*0.2

    = 500.025



    86.435T 27653 2005665 651.942 Tue 14-Oct-2003 0.864 has 1 PRP test performed.

    Thus; score = (n/1M ^ 2) * 125 * 0.6 * bias

    =((2005665/1000000)^2)*125*(86.435T/40T)*0.6

    = 651.95

  23. #143
    i don't understand why it has two done already but thanks for th info.

  24. #144
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    i don't understand why it has two done already but thanks for th info.
    Some candidates do already have 2 PRP tests. Typically these happen where the original PRP test failed to report within the 5 day deadline (or whatever it was then), it was re-issued, but the both ran to completion.

  25. #145
    shouldn't these values have been removed from the current sob.dat file then? I thought all the values with two tests or that had already had factors foundwere eliminated before. That is i atleast thought they were nolonger considered to be possible primes so they were off of our list and not worth sieveing at all. I don't know what dat i was using when i found that factor but are values that have been tested twice stilli n the current dat file for the sievers?

    P.S. it would be cool to get a list of next factors to enter secret active window. I'm sure it'll be made in due time though and I don't mind waiting as other things are definatly more important.

  26. #146
    I've also noticed that nothing from my lower ranges ever appears in the factors next to enter active window

  27. #147
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    If you mean for ranges p<40T, that's the case for everybody.

    As you might recall, we switched to the new scoring system at 40T. Every factor below 40T has already got their fullest points with the old scoring system.

  28. #148
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    Originally posted by Keroberts1
    I thought all the values with two tests or that had already had factors foundwere eliminated before. That is i atleast thought they were nolonger considered to be possible primes so they were off of our list and not worth sieveing at all.
    There's still a slight error rate even after having passed 2 tests. Unfortunately, one cannot easily verify the PRP test, whereas it is pretty simple to verify a found factor. As there is no or next to no performance penalty for such factors, there's no big advantage in omitting them...

  29. #149
    i believe factors in the 50-70 T range are missing too.

  30. #150
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    Originally posted by Keroberts1
    i believe factors in the 50-70 T range are missing too.
    I can see my factors for the 65T-67T range I seived. May be it is for some users only?? Can you specify it more?

  31. #151
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    i believe factors in the 50-70 T range are missing too.
    I see 5 factors in this range in your 'Factors next to enter (main) 'active window'' right now.

  32. #152
    It is my understanding that once we hit another FFT boundary we will start seeing the PRP tests takeing significantly longer to complete. Is this correct, and if it is will this be reflected in the scoring of the factors for that size tests. It makes sense ot me that this would be so since the current value of the factors is related to the amount of work it saved the main effort. Just a side though i might be completely off on this anyways.

  33. #153
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    Is this correct, and if it is will this be reflected in the scoring of the factors for that size tests.
    You are correct, however I rekon that improvements in the PRP client over time just about balance this out, so the square rule holds true enough.

    So no, I wasn't planning to changing the sieve/factor scoring.

  34. #154
    where is the ftt barrier? How much slower are the tests going ot run. I didn't think that the new client was any faster was any faster. I actually thought i heard it was slower. We'll see how the general public feels about this and see how the client performs o nthe new tests before making any rash decisions I'm sure.

  35. #155
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    FFT crossover points are discussed here

    When I talk about imrovemnts in the PRP client, I'm not just talking about the last release (which indeed is no quicker), but the whole history of the project, and the future.

    I did think about the FFT crossover point when the current sieve/factor scores were designed, but all this would have done would be to make the scoring use (say) n^2.1 instead of n^2. And the effect would be the same for everyone, so relative scores wouldn't change.

    In the end there are lots of different ways we could score, but this is the one we went with, and it's one with which most people seem reasonably happy.

  36. #156
    ok good en0ugh explanation for me.

  37. #157
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    Congratulations to dmbrubac for taking the number one spot in the 14 day average. Very much deserved since you are generating 0.8 'active' factors per day from factoring alone.

  38. #158
    I love 67607
    Join Date
    Dec 2002
    Location
    Istanbul
    Posts
    752
    Originally posted by MikeH
    Congratulations to dmbrubac for taking the number one spot in the 14 day average. Very much deserved since you are generating 0.8 'active' factors per day from factoring alone.
    That 0.8 figure is incredible. I envy it.

  39. #159
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    ....and now I'm 4th.

    Code:
      1+ dmbrubac                      20396.61 (12.35)     0.8 ( 1.37)   0.0      0.0      0.8     0.0      0.0 
      2+ Mystwalker                    18767.07 (11.36)     2.5 ( 4.36)   0.1      0.1      2.4     0.0      0.0 
      3+ OrkunBanuTST (Nuri)           17880.26 (10.82)    11.4 (19.80)   0.1      1.1     10.2     0.0      0.0 
      4- MikeH                         17832.56 (10.79)    11.6 (20.17)   0.3      1.1     10.4     0.0     12.5
    congratulations Nuri and Mystwalker

  40. #160
    could I have a number of factors over 1000000 that have been found that are over 40T? Just wondering to see how much further wewould have to push sieving to save another 5% of tests. I figured that since 40T sieving averages about a 5% speedup of the rate that the PRP line moves.

Page 4 of 5 FirstFirst 12345 LastLast

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •