Re: Anybody still have a test ham/spam database?
Tim Peters <[email protected]> Tue, 10 Jul 2018 21:57:24 -0500
| Newsgroups | gmane.mail.spam.spambayes.devel |
|---|---|
| Message-ID | <CAExdVNkBvz64uu4c7ORNw4Uos=0n+hSY58oucxY8k7x5M7pQHg@mail.gmail.com> |
--===============3332369772362338354== Content-Type: multipart/alternative; boundary="000000000000a720e50570b06883" --000000000000a720e50570b06883 Content-Type: text/plain; charset="UTF-8" [Skip Montanaro] > > Sure, but constructing a suitable ham/spam corpus > from scratch is a non-trivial task, as you no doubt > remember. Ah - but we had a much subtler task then: trying to construct a classifier that was _useful_. Your current task is much clearer: > ... I am looking to insure that a Py3 port of SpamBayes > works the same as the Py2 code. For _that_ purpose, you can take any pile of email at all; split it into "ham" and "spam" at random, and "just" ensure you get the same results from the older and newer code. Your criterion for success isn't "closeness to human value judgment", but "same output". For that purpose, you could synthesize gibberish email from random header & sentence generators. Although it would be easier to use real email ;-) The point is that you don't have to worry at all about whether this or that is "really ham" or "really spam" or "really unsure" - it was making those value judgments that consumed lots of human time when building the old curated data sets. --000000000000a720e50570b06883 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable <div dir=3D"ltr"><br><div class=3D"gmail_quote"><div dir=3D"ltr">[Skip Mont= anaro]</div><blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;bo= rder-left:1px #ccc solid;padding-left:1ex"></blockquote>> Sure, but cons= tructing a suitable ham/spam corpus<br>> from scratch is a non-trivial t= ask, as you no doubt<br>> remember.</div><div class=3D"gmail_quote"><br>= </div><div class=3D"gmail_quote">Ah - but we had a much subtler task then:= =C2=A0 trying to construct a classifier that was _useful_.=C2=A0 Your curre= nt task is much clearer:</div><div class=3D"gmail_quote"><br></div><div cla= ss=3D"gmail_quote">> ... I am looking to insure that a Py3 port of SpamB= ayes<br>> works the same as the Py2 code.<br><div><br>For _that_ purpose= , you can take any pile of email at all; split it into "ham" and = "spam" at random, and "just" ensure you get the same re= sults from the older and newer code.=C2=A0 Your criterion for success isn&#= 39;t "closeness to human value judgment", but "same output&q= uot;.<br><br>For that purpose, you could synthesize gibberish email from ra= ndom header & sentence generators.=C2=A0 Although it would be easier to= use real email ;-)=C2=A0 The point is that you don't have to worry at = all about whether this or that is "really ham" or "really sp= am" or "really unsure" - it was making those value judgments= that consumed lots of human time when building the old curated data sets.<= br><br></div></div></div> --000000000000a720e50570b06883-- --===============3332369772362338354== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ spambayes-dev mailing list [email protected] https://mail.python.org/mailman/listinfo/spambayes-dev --===============3332369772362338354==--