2003-10-24, 06:53 | #1 |
Sep 2003
3×863 Posts |
[Edit: this thread has now been split off from the original "Which exponents should be re-released for first time tests?" thread. The original list of 498 exponents is given as an attachment within that thread (look for the file 498.zip).]
There is a new release of data files dated Oct 24. So here's an update on the status of the exponents. There were 498 exponents released (489 required a 2nd test, and 9 required a 3rd test because both prior results were by error-prone machines). There were also 67 exponents that would have been released according to the criteria, but were already assigned (all 67 required a 2nd test, none required a 3rd test or higher). Of the 9 triple-checks, 6 are still pending and 3 have completed, all now verified good. Closer inspection of the shift counts and residues shows that these 3 weren't really triple-checks at all; rather machines S62101/C7B7B3D04 and pbtek23/pbtek23 might be the same machine. Of the 489 double-check exponents, 428 are still pending, 17 now require triple-checks (original test therefore presumed bad) and 44 are now verified good. This is an error rate of 17/(17+44) = 0.278, roughly in line with the 0.333 error rate which was the criterion for detecting error-prone machines. Of the 67 already-assigned double-check exponents (not part of the 498 that were released), 49 are still pending, 10 now require triple-checks (original test therefore presumed bad) and 8 are now verified good. This is an error rate of 10/(10+8) = 0.555. More details are in the attached file. Last fiddled with by GP2 on 2003-11-17 at 22:28 |
2003-11-02, 08:29 | #2 |
Sep 2003
3×863 Posts |
There is a new release of data files dated Oct 31. So here's an update on the status of the exponents.
There were 498 exponents released (489 required a 2nd test, and 9 required a 3rd test because both prior results were by error-prone machines). There were also 67 exponents that would have been released according to the criteria, but were already assigned (all 67 required a 2nd test, none required a 3rd test or higher). Of the 9 triple-checks, 5 are still pending and 3 are now confirmed good, and 1 now needs a quadruple check. Of the 489 (out of 498) exponents released for a 2nd test, 386 exponents are still assigned, 2 exponents very recently cleared, 31 now require triple-checks (the original test is therefore presumed bad) and 70 are now verified good. This is an error rate of 31/(31+70) = 0.306, roughly in line with the 0.333 error rate which was the criterion for detecting error-prone machines. Of the 67 already-assigned double-check exponents (not part of the 498 that were released), 39 are still pending, 5 have apparently expired and not been reassigned, 15 now require triple-checks (the original test is therefore presumed bad) and 8 are now verified good. This is an error rate of 15/(15+8) = 0.652. More details are in the attached file. Last fiddled with by GP2 on 2003-11-02 at 08:34 |
2003-11-02, 15:03 | #3 |
Aug 2002
Texas
9B_{16} Posts |
GP2,
I just want to commend you on all the great data mining you do around here. You have been a great addition to the GIMPS family and provide, at least to me, very interesting information. Gratuitous dancing |
2003-11-02, 20:50 | #4 |
Sep 2003
3×863 Posts |
Thanks Complex33.
Looking at the preliminary results, we see a significantly higher error rate from the exponents that "would have been released" under the error-prone-machine criteria, except they were already assigned at the time. This makes sense, because the exponents that had already been assigned were for the most part those with "harmful" error codes, which statistically indicates a 55%-60% chance of an erroneous result. George had already released these earlier. Whereas, for the exponents that were actually released, the criteria were a 33% error rate for the error-prone machines. So you'd expect a similar error rate. Out of the 31 exponents that now require triple-checks (probably indicating that the original test was bad, although it's possible the 2nd test could be the bad one, or even both), 24 of them now have both results with 00000000 error code (at least one of which, by definition, is wrong). So these represent errors which would not have been caught by relying on "harmful" error codes alone. So I think this release of exponents has been a modest success in terms of the goal of early re-testing of exponents which may not have had a correct first-time LL test. If there's general agreement, I can propose a new batch of exponents for release, with some slightly expanded criteria. |
2003-11-10, 02:07 | #5 |
Sep 2003
3×863 Posts |
The status as of 2003-11-09 19:00 UTC:
Of the 9 (out of 498) exponents released for a 3rd or higher test (because both original tests were by error-prone machines), 5 are still assigned, 3 are now confirmed good, 1 now needs a quadruple check. Of the 489 (out of 498) exponents released for a 2nd test, 355 exponents are still assigned, 2 have been factored, 2 exponents very recently cleared, 38 now require triple-checks (the original test is therefore presumed bad) and 92 are now verified good. This is an error rate of 38/(38+92) = .292, roughly in line with the 0.333 error rate which was the criterion for detecting error-prone machines. 67 exponents fit the criteria but were not part of the file_count exponents because they were already assigned. Of these, all 67 exponents had been tested once and needed a 2nd test. Of these 67 exponents, 37 are still assigned, 5 had apparently expired but are today being reassigned as part of the second release of exponents, 16 now require triple-checks (the original test is therefore presumed bad) and 9 are now verified good. This is an error rate of 16/(16+9) = .640. More details in the attached file: Last fiddled with by GP2 on 2003-11-10 at 02:45 |
2003-11-17, 22:34 | #6 |
Sep 2003
3×863 Posts |
The status as of 2003-11-17 20:00 UTC:
Of the 9 (out of 498) exponents released for a 3rd or higher test (because both original tests were by error-prone machines), 5 are still assigned, 3 are now confirmed good, 1 now needs a quadruple check. Of the 489 (out of 498) exponents released for a 2nd test, 334 exponents are still assigned, 2 have been factored, 0 have apparently expired and not been reassigned, 4 exponents very recently cleared, 44 now require triple-checks (the original test is therefore presumed bad) 105 are now verified good. This is an error rate of 44/(44+105) = .295, roughly in line with the 0.333 error rate which was the criterion for detecting error-prone machines. 67 exponents fit the criteria but were not part of the 498 exponents because they were already assigned. Of these, 67 exponents had been tested once and needed a 2nd test. Of these 67 exponents, 42 are still assigned, 0 have been factored, 0 have apparently expired and not been reassigned, 15 now require triple-checks (the original test is therefore presumed bad) 9 are now verified good. 1 is now verified bad after a quadruple check. This is an error rate of 1+15/(1+15+9) = .640. More details in the attached file: |
2003-12-03, 04:57 | #7 |
Sep 2003
3×863 Posts |
The status as of 2003-12-03 2:00 UTC:
Of the 9 (out of 498) exponents released for a 3rd or higher test (because both original tests were by error-prone machines), 5 are still assigned, 3 are now confirmed good, 1 now needs a quadruple check. Of the 489 (out of 498) exponents released for a 2nd test, 306 exponents are still assigned, 2 have been factored, 0 have apparently expired and not been reassigned, 5 exponents very recently cleared, 49 now require triple-checks (the original test is therefore presumed bad) 127 are now verified good. This is an error rate of 49/(49+127) = .278, roughly in line with the 0.333 error rate which was the criterion for detecting error-prone machines. 67 exponents fit the criteria but were not part of the 498 exponents because they were already assigned. Of these, 67 exponents had been tested once and needed a 2nd test. Of these 67 exponents, 36 are still assigned, 0 have been factored, 0 have apparently expired and not been reassigned, 17 now require triple-checks (the original test is therefore presumed bad) 13 are now verified good. 1 is now verified bad after a quadruple check. This is an error rate of 1+17/(1+17+13) = .581. More details in the attached file: |
2003-12-15, 19:16 | #8 |
Sep 2003
3·863 Posts |
The status as of 2003-12-14 17:00 UTC:
Of the 9 (out of 498) exponents released for a 3rd or higher test (because both original tests were by error-prone machines), 4 are still assigned, 3 are now confirmed good, 1 now needs a quadruple check, 1 has recently cleared. Of the 489 (out of 498) exponents released for a 2nd test, 291 exponents are still assigned, 2 have been factored, 2 have apparently expired and not been reassigned, 3 exponents very recently cleared, 54 now require triple-checks (the original test is therefore presumed bad) 137 are now verified good. This is an error rate of 54/(54+137) = .282, roughly in line with the 0.333 error rate which was the criterion for detecting error-prone machines. 67 exponents fit the criteria but were not part of the 498 exponents because they were already assigned. Of these, 67 exponents had been tested once and needed a 2nd test. Of these 67 exponents, 29 are still assigned, 0 have been factored, 0 have apparently expired and not been reassigned, 22 now require triple-checks (the original test is therefore presumed bad) 15 are now verified good. 1 is now verified bad after a quadruple check. This is an error rate of (1+22)/(1+22+15) = .605. |
2004-03-18, 02:46 | #9 |
Nov 2003
3×5×11 Posts |
I thought I'd update this status page, as the data could be useful. Note that all error rate calculations assume the original test was bad.
The status as of 2004-Mar-18 2:00 UTC: Of the 9 (out of 498) exponents released for a 3rd test (because both original tests were by error-prone machines), 1 is still pending, 1 needs a quadruple check, 7 now have confirmed results, with only 1 erroneous result. Of the 489 (out of 498) exponents released for a 2nd test, 56 are still assigned, 4 have been factored, 1 have apparently expired and not been reassigned, 0 have very recently cleared, 321 have been confirmed good. The remaining 107 exponents have an erroneous result. (For calculating error rates the original test is presumed to be the bad one). 15 have had two matching results, and one bad result. 11 triple checks are assigned by primenet. The remaining 81 are probably waiting to be assigned as a normal double check. This is an error rate of 107/(107+321)=.250, slightly lower that the 0.333 error rate which was the criterion for detecting error prone machines. 67 exponents fit the criteria but were not part of the 498 exponents because they were already assigned. Of these, 67 exponents had been tested once and needed a 2nd test. Of these 67 exponents, 13 are still assigned, 0 have been factored, 0 have apparently expired and not been reassigned, 17 are now verified good. The remaining 37 have erroneous results. 10 exponents have two good results and one bad result, 1 exponent has two good results, and two bad results, 1 triple check is pending on primenet, 25 others still require triple-checks. This is an error rate of (1+37)/(1+37+17)=.691 More detail in the attached file: |
2004-08-12, 21:18 | #10 |
Nov 2003
3×5×11 Posts |
The status as of 2004-Aug-12 20:00 UTC:
Of the 9 (out of 498) exponents released for a 3rd test (because both original tests were by error-prone machines), 1 has a quadruple check on primenet, 1 needs a quadruple check, 6 have three matching results, 1 has 2 matching results and 1 erroneous result. Of the 489 (out of 498) exponents released for a 2nd test, 10 are still pending, 4 have been factored, 0 have apparently expired and not been reassigned, 1 has very recently cleared, 358 have been confirmed good. The remaining 116 exponents have an erroneous result. (For calculating error rates the original test is presumed to be the bad one). 47 have two matching results, and one bad result, 1 has two matching results, and two bad results, 36 triple checks are assigned by primenet, 1 needs a quadruple check, 31 need a triple check. This is an error rate of 116/(116+358)=.245, slightly lower that the 0.333 error rate which was the criterion for detecting error prone machines. 67 exponents fit the criteria but were not part of the 498 exponents because they were already assigned. Of these, 67 exponents had been tested once and needed a 2nd test. Of these 67 exponents, 6 are still assigned, 0 have been factored, 0 have apparently expired and not been reassigned, 17 are now verified good. The remaining 44 have erroneous results. 17 exponents have two good results and one bad result, 1 exponent has two good results, and two bad results, 2 triple checks are assigned by primenet, 1 needs a quadruple check 23 need a triple check. This is an error rate of (44)/(44+17)=.721 More detail in the attached file: |
2005-07-25, 13:14 | #11 |
Aug 2002
Termonfeckin, IE
2764_{10} Posts |
I thought this thread needs an update.
The status as of 2005-Jul-25 12:00 UTC: I did not track triple and quadruplechecks but here is a short summary: Of the 498 exponents released for a 2nd/3rd test, 4 have been factored, 1 expired and was not reassigned. 370 have been confirmed good. The remaining 123 exponents have an erroneous result. (For calculating error rates the original test is presumed to be the bad one). 1 has been cleared and needs to be reassigned for quadruplechecking, 97 have two matching results, and one bad result, 5 have two matching results, and two bad results, 1 has two matching results, and three bad results, 3 triple checks are assigned by primenet, 16 need a triplecheck. This is an error rate of 123/(123+370)=.249, slightly lower that the 0.333 error rate which was the criterion for detecting error prone machines. Last fiddled with by garo on 2005-07-25 at 13:17 |
Thread Tools | |
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Tracking the November 2003 release of error-prone exponents | nfortino | Data | 2 | 2005-07-25 13:46 |
2003 Nov 03: P-1: a set of 30 exponents | GP2 | Completed Missions | 3 | 2003-11-17 15:09 |
2003-10-29: P-1: a set of 26 larger exponents | GP2 | Completed Missions | 3 | 2003-11-12 14:16 |
2003 Oct 31: P-1: a set of 17 larger exponents | GP2 | Completed Missions | 2 | 2003-11-03 15:34 |
107 stuck exponents at 20.8M as of Oct 24 2003 | GP2 | Data | 1 | 2003-10-24 21:32 |