From spamassassin-talk-admin@lists.sourceforge.net  Wed Aug 21 13:02:24 2002
Return-Path: <spamassassin-talk-admin@example.sourceforge.net>
Delivered-To: yyyy@localhost.netnoteinc.com
Received: from localhost (localhost [127.0.0.1])
	by phobos.labs.netnoteinc.com (Postfix) with ESMTP id 2078B43C32
	for <jm@localhost>; Wed, 21 Aug 2002 08:02:24 -0400 (EDT)
Received: from phobos [127.0.0.1]
	by localhost with IMAP (fetchmail-5.9.0)
	for jm@localhost (single-drop); Wed, 21 Aug 2002 13:02:24 +0100 (IST)
Received: from usw-sf-list2.sourceforge.net (usw-sf-fw2.sourceforge.net
    [216.136.171.252]) by dogma.slashnull.org (8.11.6/8.11.6) with ESMTP id
    g7LC02Z22978 for <jm-sa@jmason.org>; Wed, 21 Aug 2002 13:00:02 +0100
Received: from usw-sf-list1-b.sourceforge.net ([10.3.1.13]
    helo=usw-sf-list1.sourceforge.net) by usw-sf-list2.sourceforge.net with
    esmtp (Exim 3.31-VA-mm2 #1 (Debian)) id 17hU7f-00050W-00; Wed,
    21 Aug 2002 04:58:03 -0700
Received: from cs.rice.edu ([128.42.1.30]) by usw-sf-list1.sourceforge.net
    with esmtp (Exim 3.31-VA-mm2 #1 (Debian)) id 17hU7H-0001Ok-00 for
    <spamassassin-talk@lists.sourceforge.net>; Wed, 21 Aug 2002 04:57:39 -0700
Received: from localhost (localhost [127.0.0.1]) by cs.rice.edu (Postfix)
    with ESMTP id 801104AA47; Wed, 21 Aug 2002 06:57:38 -0500 (CDT)
Received: from bert.cs.rice.edu (bert.cs.rice.edu [128.42.3.146]) by
    cs.rice.edu (Postfix) with ESMTP id 90CFD4AA43; Wed, 21 Aug 2002 06:57:37
    -0500 (CDT)
Received: by bert.cs.rice.edu (Postfix, from userid 14314) id 7A930374003;
    Wed, 21 Aug 2002 06:57:36 -0500 (CDT)
To: "Craig R.Hughes" <craig@deersoft.com>
Cc: yyyy@spamassassin.taint.org (Justin Mason),
	"Malte S. Stretz" <msquadrat.nospamplease@gmx.net>,
	SpamAssassin Talk ML <spamassassin-talk@lists.sourceforge.net>
References: <B7973337-B468-11D6-9F04-00039396ECF2@deersoft.com>
From: Scott A Crosby <scrosby@cs.rice.edu>
Organization: Rice University
In-Reply-To: <B7973337-B468-11D6-9F04-00039396ECF2@deersoft.com>
Message-Id: <oydelcsv2z4.fsf@bert.cs.rice.edu>
User-Agent: Gnus/5.0808 (Gnus v5.8.8) XEmacs/21.4 (Common Lisp)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Virus-Scanned: by AMaViS snapshot-20020300
Subject: [SAtalk] Re: results for giant mass-check (phew)
Sender: spamassassin-talk-admin@example.sourceforge.net
Errors-To: spamassassin-talk-admin@example.sourceforge.net
X-Beenthere: spamassassin-talk@example.sourceforge.net
X-Mailman-Version: 2.0.9-sf.net
Precedence: bulk
List-Help: <mailto:spamassassin-talk-request@example.sourceforge.net?subject=help>
List-Post: <mailto:spamassassin-talk@example.sourceforge.net>
List-Subscribe: <https://example.sourceforge.net/lists/listinfo/spamassassin-talk>,
    <mailto:spamassassin-talk-request@lists.sourceforge.net?subject=subscribe>
List-Id: Talk about SpamAssassin <spamassassin-talk.example.sourceforge.net>
List-Unsubscribe: <https://example.sourceforge.net/lists/listinfo/spamassassin-talk>,
    <mailto:spamassassin-talk-request@lists.sourceforge.net?subject=unsubscribe>
List-Archive: <http://www.geocrawler.com/redir-sf.php3?list=spamassassin-talk>
X-Original-Date: 21 Aug 2002 06:57:35 -0500
Date: 21 Aug 2002 06:57:35 -0500

On Tue, 20 Aug 2002 11:14:50 -0700, Craig R.Hughes <craig@deersoft.com> writes:

> On Tuesday, August 20, 2002, at 07:23  AM, Justin Mason wrote:
> 
`> Remember that the GA is going to be considering combinatorial uses of
> the rules, so rules which look dodgy on their own might be gems for
> the GA -- perhaps something with a S/O ratio of .5 actually occurs
> often in combination with some other rule, and in those situations,
> helps to distinguish spam vs nonspam.
> 

We're currently using a perceptron classifier. It *can't* learn
combinations of rules.[1]

I gave an example, assume 4 rules:
   SPAM = (A or B) and (C or D)

It cannot learn that function.

A decision tree classifier *can* learn that example, and the function
above, where a .5 S/O rule is only important in certain
circumstances. (Then again, it may be smartest to hardcode a meta-rule
for that case, rather than trust to a naive DT learner.)

Scott


[1] To be fair, nor can a Bayes classifier.


-------------------------------------------------------
This sf.net email is sponsored by: OSDN - Tired of that same old
cell phone?  Get a new here for FREE!
https://www.inphonic.com/r.asp?r=sourceforge1&refcode1=vs3390
_______________________________________________
Spamassassin-talk mailing list
Spamassassin-talk@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk