>> Does Spambayes learn from experience or only from training? >If you mean does it do any automatic training, then no it does not.
No what I meant was if I go into my Junk suspects folder and delete as spam and recover from spam doesn't spambayes learn from that? Additionally I was reading about ways to clean up the database. Can I open my database of messages it trained on and delete enough of the HAM so it gets closer to an even match. I am still discussing it because that one oddball false positive is still occurring. It is exactly the same content as the messages I have sent several times. The score is 0% in the Outlook column yet it repeatedly drops it into my spam folder. What I find even odder is that I have a rule that works for 99 out of 100 such messages. How these even slip by the rule is just wildly confusing to me. I have searched through my filters and I have no filters that send mail to the Junk folder. Additionally I run no other anti spam software and have Outlook junkmail disabled. I am aware that this does not mean that spambayes is the cause of this issue but since it is the only program I believe set to filter mail to the Junk E-mail folder I am guessing it is related. Based on the fact that you clearly say smaller database of more equal amounts of ham and spam messages are better than the odd imbalance I have plus the fact that I am getting even this one anomaly I want to fix that imbalance. If I can edit the database that seems to be the best choice, if not I will retrain. But if I just open the manager and retrain does it simply recreate the database based on the new data or does it add to the current database? Thanks for the help Doug -----Original Message----- From: Tony Meyer [mailto:[EMAIL PROTECTED] Sent: Friday, June 24, 2005 3:46 AM To: 'douglas cohn' Cc: [email protected] Subject: RE: [Spambayes] Odd behavior with messages > Basically I have been saving my spam for sometime based on your > statement and some others. I have tons of good mail and wanted to > give it corss sections to train on. What I could do is copy a few > messages from each folder of the good mail and then train on it. I > have a lot of log reports and status emails so training is very > important and I must say none of those messages get caught. I also > get very few suspects these days. > > If it is working well must I retrain? No. The golden rule is that if it's working then you should leave it as is. I'm just warning that you'll see some really strange scores with such a huge imbalance. All testing has shown that training regimes like 'train on error' (i.e. false positives, false negatives and unsures) do much better than bulk training, and that generally smaller databases are more accurate than large ones. > Does Spambayes learn from experience or only from training? If you mean does it do any automatic training, then no it does not. =Tony.Meyer -- Please always include the list (spambayes at python.org) in your replies (reply-all), and please don't send me personal mail about SpamBayes. http://www.massey.ac.nz/~tameyer/writing/reply_all.html explains this. _______________________________________________ [email protected] http://mail.python.org/mailman/listinfo/spambayes Check the FAQ before asking: http://spambayes.sf.net/faq.html
