training to exhaustion?

Tom Allison tallison at tacocat.net
Tue Mar 9 13:35:10 CET 2004


David Relson wrote:
> On Tue, 09 Mar 2004 06:11:03 -0500
> Tom Allison wrote:
> 
> 
>>I've been doing the following and have one question.
>>How many times to train the whole list?  I've on iteration number 6 
>>right now and I have two emails that are still coming up spam.
>>
>>Another question:
>>These have all been filtered once, should I kill the 'N' option after 
>>the first time through since they were only mis-read once?  (I'm 
>>guessing probably?)
> 
> 
> Yes.  Personally, I don't "train until right".  I figure that, over
> time, bogofilter will learn what it needs...
> 

Interestingly, some of the more difficult to train where actually 
misclassified by me.  Fixing that, and reloading a backup wordlist, I 
think it's working.  However, since all the email I was processing was 
nothing to be "unlearned" I skipped the -S -N options.

Two passes and it's done.

Now we'll see how it goes.

Made some cute scripts for it though.  I think I'll run a monthly test 
on my archives to see how many of them have changed state based on 
current learning patterns.





More information about the Bogofilter mailing list