[Spambayes] interacting with dbm

[Luca Benassi]

Jesse Pelton ha scritto:
> Ah.  Perhaps the thing to do is to build a starting database by training
> on a corpus of ham and spam messages that you know your clients would
> agree that you've classified correctly.  You can then install that
> database on their machines when you install SpamBayes.  See
> http://spambayes.sourceforge.net/faq.html#can-i-share-move-my-training-d
> ata-from-one-computer-to-another.

Yeah! That's it!
Thx also to Amedee.

I'll try this out, thx again.

> -----Original Message-----
> From: Luca Benassi [mailto:benassi at labs.it] 
> Sent: Thursday, March 08, 2007 8:22 AM
> To: Jesse Pelton
> Cc: spambayes at python.org
> Subject: Re: [Spambayes] interacting with dbm
> 
> Jesse Pelton ha scritto:
>> As I understand it, the thing to do is to train on messages that
> contain
>> those words.  That is, if you receive a message that contains those
>> words and SpamBayes classifies it as spam or possible spam, tell
>> SpamBayes that it's actually ham.  Alternatively, you could probably
>> parse through the code to understand the database design and hack up
>> something to add some special tokens to the database.  Heck, you could
>> change the code itself to give special treatment to your whitelist
>> tokens.  The source is readily available.
>>
>> I don't recommend any of the above, though.  If you somehow receive a
>> spam message with one of your whitelist words in it, the first
> approach
>> would require you to mislead SpamBayes about what you consider to be
>> spam, and the others seem very error-prone.
>>
>> Lots of people have tried to find ways to make SpamBayes smarter.  As
>> near as I can tell, no one has found a way to make it consistently
>> perform better in a long time.  Does it matter that it's dumb if it
>> gives good results?
> 
> thx for your reply ... you're right ;)
> however, what I'd like to try is to pre-train spambayes, before (or, 
> *while*) installing to some clients.

> 
> 
>> -----Original Message-----
>> From: spambayes-bounces at python.org
> [mailto:spambayes-bounces at python.org]
>> On Behalf Of Luca Benassi
>> Sent: Thursday, March 08, 2007 6:42 AM
>> To: spambayes at python.org
>> Subject: [Spambayes] interacting with dbm
>>
>> I've serched a bit in doc, faq and old messagges but ... no luck :]
>>
>> This is my question: can I populate the dbm with my list of *safe* 
>> (good) words? I'd like to, sort of, whitelist a group of words.
>>
>> Thx in advice,
>> Luca
>>
> 
> 


-- 
Dr. Luca Benassi
Laboratori Guglielmo Marconi
Via Porrettana 123, 40037 Pontecchio Marconi (BO) - ITALY
Phone:+39-0516781934 Fax:+39-051846479 e-mail: benassi at labs.it
Systems & Networks Division