Re: interesting cssdiff output
<[email protected]> Tue, 15 Jun 2010 09:39:25 -0400
| Newsgroups | gmane.mail.spam.crm114 |
|---|---|
| Message-ID | <[email protected]> |
From: Jason White <[email protected]> I have just started training CRM114 for spam detection (when last I tried it a few years ago, I wasn't able to achieve high accuracy and deleted the CSS files). This time, I'm running it directly from the Debian package, version 20090807, x86-64 architecture with an up to date Debian Sid system. It's too early to judge accuracy, since I haven't received enough spam in the last day or so to give sufficient training. However, I notice the following suspect numbers in the cssdiff output: cssdiff nonspam.css spam.css Sparse spectra file nonspam.css has 1048577 bins total Sparse spectra file spam.css has 1048577 bins total File 1 total features : 16547 File 2 total features : 7219 Similarities between files : 665 Differences between files : 14081050284681 File 1 dominates file 2 : 47244656123 File 2 dominates file 1 : 734439413993 The difference and dominance lines seem highly suspect to me, either due to a problem with the calculations or corrupt CSS files. Alternatively, this could be correct and I'm just making unwarranted assumptions about the numbers. Whoa. Yeah, that's broken. I wonder if it's due to the 64-bit-isms not having been taken care of adequately in the 2009 release. The trick is we're shifting over to the ANSI C library form (which _is_ scrubbed 64-bit-clean, thanks Kurt!) and will then backport that into the standalone language. In that case, the bug is in code that's going away anyway. If the classifiers are working right for you (where that means "accuracy is improving and, within a week or so, satisfactory or better) then I'll just peg this as a "will be fixed in next release". Otherwise, I'll work the bug as a P2 at least. - Bill Yerazunis ------------------------------------------------------------------------------ ThinkGeek and WIRED's GeekDad team up for the Ultimate GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the lucky parental unit. See the prize list and enter to win: http://p.sf.net/sfu/thinkgeek-promo