[Spambayes] Problem with POP3 Proxy: Sorting Accuracy?

bill at antietamtravel.com bill at antietamtravel.com
Wed May 5 09:53:11 EDT 2004


A non-text attachment was scrubbed...
Name: SpamBayesServer1.log
Type: application/octet-stream
Size: 1374 bytes
Desc: not available
Url : http://mail.python.org/pipermail/spambayes/attachments/20040505/1635a1f9/SpamBayesServer1.obj
-------------- next part --------------
I am using SpamBayes POP3 Proxy Version 1.0b1 (April 2004) (binary),
with version 2.3.3 (#51, Feb 13 2004, 14:39:56) [MSC v.1200 32 bit
(Intel)] of Python; my operating system is Windows 5.0.2195.2 (Service
Pack 4).  I have trained 568 ham and 2412 spam. I am using Eudora 5.1
as my email client
 
The problem I am having is with sorting. I have been using Spambayes
for about a week now. As you can see from the statistics below, I
receive a significant amount of spam, ~81%.
 
After a week, I feel that Spambayes is not sorting accurately. I have
trained accurately, but this morning, all of the deferred messages -
30 or 40 - were spam. About 20 of 60 messages sorted as ham were, in
fact, spam. The messages sorted as spam are indeed spam, but Spambayes
is missing too many.
 
Am I doing something wrong? I just saw a post that someone was getting
99% accuracy after only 2 days! Any suggestions you can provide will
be greatly appreciated.
 
Status and Configuration: 
POP3 proxy running on 1110, 1120, proxying to
mail.antietamtravel.com:110, mail.adelphia.net:110.
Active POP3 conversations: 0. 
POP3 conversations this session: 306. 
Emails classified this session: 427 spam, 276 ham, 131 unsure. 
Total emails trained: Spam: 2412 Ham: 568 
 
Statistics: 
 
SpamBayes has processed 2980 messages - 1031 (35%) good, 1319 (44%)
spam and 630 (21%) unsure.
 
568 messages were manually classified as good (0 were false
positives).
 
2412 messages were manually classified as spam (482 were false
negatives).
 
19 unsure messages were manually identified as good, and 611 as spam.
 
Thanks, 
 
Bill


More information about the Spambayes mailing list