Re: SA Question

[Joe Emenaker]

Richard Whittaker wrote:

Actually, I figured out a means of using the IMAP/Folder method, and it
works great, and doesn't require much care and feeding by me (which is
really what mattered!)...

What I've recently done is made two IMAP folders called "Spam/LearnSpam" and "Spam/LearnHam". I've got all autolearning turned OFF since, if the normal rules give a false-negative or a false-positive, then I suspect that the Bayes component will start to re-inforce it. In other words, it would just "amplify" the scores being given by the other rules, in a sense.

So, once I've got those two folders for my trash, I drag all ham/spam from my inbox into the appropriate folders. Then, every hour, I have a script run (which I'll include at the end of this). Basically, what the script does is take everything from LearnSpam and runs "sa-learn --spam" on it and does the same kinda deal with LearnHam.

One additional thing that it does is it grabs the spam-score from each message and writes it to a spamlog file (along with whether it found it in the spam-trash or the good-trash). This way, I can do a "sort -n" on the file and immediate tell:
1 - If there's *any* overlap between the scores on the spam and the ham. If there's overlap, then there's no way to eliminate false-positives AND false-negatives without modifying the rules or the having the bayes db learn more.
2 - If there is *no* overlap, then it tells you how big of a gap you have between your lowest-scored spam and your highest-scored ham. The idea here is that you could set your required_hits to be right in the middle of this gap.

- Joe

--- learnspam.sh ---

#!/bin/sh

MAILDIR="$HOME/Maildir"
HAMDIR="$MAILDIR/.Spam.LearnHam"
SPAMDIR="$MAILDIR/.Spam.LearnSpam"
SPAMLOG="$HOME/spamlog"

record() {
score=`cat "$2" | perl -ne 'print if(s/^X-Spam-Status: .*hits=([-.0-9]+) required.*/$1/);' | head -1`
if [ -n "$score" ]
then
echo "$score : $1" >> "$SPAMLOG"
fi
rm "$2"
}

for name in `find "$SPAMDIR" -type f -name "*,S=*"`
do
        sa-learn --spam "$name"
        record "SPAM" "$name"
done

for name in `find "$HAMDIR" -type f -name "*,S=*"`
do
        sa-learn --ham "$name"
        record "GOOD" "$name"
done

smime.p7s

Description: S/MIME Cryptographic Signature