Updating GNFS parameters

bchaffin · 2011-06-21, 16:44

Just cross-posting an advertisement in case anyone here doesn't follow the factoring forum. I'm starting a data collection effort to update the default GNFS parameters for larger numbers.

Folks working on Aliquot sequences probably factor lots of medium-to-large composites with GNFS, so hopefully you all have (or will have) lots of data for me.

Thanks!

Batalov · 2011-06-21, 16:56

M.Kamada's site has had everything you would possibly need - for years.

P.S. I agree, Tom. What I left unsaid was that graphs are only a tip of the iceberg; the site has all particulars in the report files for each (well, most) jobs. Everything can be screenscraped, the outliers (and the outlier contributors) can be flagged and then there will still be enough stats left for a decent analysis in the low range which is apparently of interest for most (aliquot) people. For high ranges, we can easily agree that using off-the-shelf scripts is admittedly silly. Then again, "high" is different for every individual.

fivemack · 2011-06-21, 17:11

I'm not quite sure that's true: it's not splitting out the number-of-relations figures by lpbr and by siever used, which are I believe important terms in the model, and a lot of it seems to be self-reflecting (eg Domanov is clearly using the linear fit of rlim against digit count to decide which rlim to use)

bchaffin · 2011-06-22, 21:19

Quote:

Originally Posted by Batalov

M.Kamada's site has had everything you would possibly need - for years.

P.S. I agree, Tom. What I left unsaid was that graphs are only a tip of the iceberg; the site has all particulars in the report files for each (well, most) jobs. Everything can be screenscraped, the outliers (and the outlier contributors) can be flagged and then there will still be enough stats left for a decent analysis in the low range which is apparently of interest for most (aliquot) people. For high ranges, we can easily agree that using off-the-shelf scripts is admittedly silly. Then again, "high" is different for every individual.

Thanks, that's some great data. I looked around on the site but couldn't find any log files for individual jobs -- can you point me at them?

Batalov · 2011-06-23, 03:33

They are stored behind the link Contributions, e.g. by date. Or by series.
Example.

Andi47 · 2011-06-23, 07:51

Quote:

Originally Posted by Batalov

They are stored behind the link Contributions, e.g. by date. Or by series.
Example.

These seem to contain the search range, but not the number of relations. (the logs in your link say "relations: relations" with the actual number of rels missing.)

I have uploaded the logs of 332 GNFS factorizations (zipped, from aliquot search, done with FactMsieve.pl), ranging from c82 to c145 to sendspace: http://www.sendspace.com/file/i8h260

bchaffin · 2011-06-23, 17:36

The logs on Kamada's site are not uniform; they probably depend on what tool/version was used for the factorization and whether the steps were performed manually or by the tool. But many of them do contain the number of relations, and also the other parameters, so they will definitely be a valuable source. I'll do some scripting to pull them all.

Thanks a lot for your logs Andi, that's great data.

2011-06-21, 16:44	#1
bchaffin Sep 2010 Portland, OR 2²×3×31 Posts	Updating GNFS parameters Just cross-posting an advertisement in case anyone here doesn't follow the factoring forum. I'm starting a data collection effort to update the default GNFS parameters for larger numbers. Folks working on Aliquot sequences probably factor lots of medium-to-large composites with GNFS, so hopefully you all have (or will have) lots of data for me. Thanks!

2011-06-21, 16:56	#2
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3·29·113 Posts	M.Kamada's site has had everything you would possibly need - for years. P.S. I agree, Tom. What I left unsaid was that graphs are only a tip of the iceberg; the site has all particulars in the report files for each (well, most) jobs. Everything can be screenscraped, the outliers (and the outlier contributors) can be flagged and then there will still be enough stats left for a decent analysis in the low range which is apparently of interest for most (aliquot) people. For high ranges, we can easily agree that using off-the-shelf scripts is admittedly silly. Then again, "high" is different for every individual. Last fiddled with by Batalov on 2011-06-21 at 17:54 Reason: added link, P.S.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Automatically updating factordb	chris2be8	FactorDB	9	2012-05-03 10:20
report not updating ???	nobody	Information & Answers	3	2012-04-26 14:31
exponent not updating problem	Nestea80	Information & Answers	3	2008-12-04 16:47
Updating server with completed P-1 runs	Prime95	Marin's Mersenne-aries	6	2003-12-30 01:52
Status Updating	SatishM	Software	4	2003-04-05 16:08

2011-06-21, 17:11	#3
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 3³·239 Posts	I'm not quite sure that's true: it's not splitting out the number-of-relations figures by lpbr and by siever used, which are I believe important terms in the model, and a lot of it seems to be self-reflecting (eg Domanov is clearly using the linear fit of rlim against digit count to decide which rlim to use)

2011-06-23, 03:33	#5
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3×29×113 Posts	They are stored behind the link Contributions, e.g. by date. Or by series. Example.

2011-06-23, 17:36	#7
bchaffin Sep 2010 Portland, OR 2²×3×31 Posts	The logs on Kamada's site are not uniform; they probably depend on what tool/version was used for the factorization and whether the steps were performed manually or by the tool. But many of them do contain the number of relations, and also the other parameters, so they will definitely be a valuable source. I'll do some scripting to pull them all. Thanks a lot for your logs Andi, that's great data.