First off, please quote the relevant portions of the message you are
replying to. Most of the users of this list use it as a mailing list
rather than a message board and cannot see the history of previous
messages unless they happen to have them saved.
On 8/4/2014 3:28 PM, RobertGrimes wrote:
First, I am running on a Windows 2003 server platform if that makes any
difference with spamd. The only parameter I use is -l.
I have tried running SA from a command line and I get significantly larger
scores than what is being generated via hMailServer. I was thinking that the
SA headers being added was confusing the command line version. If they are
stripped, why such huge differences, like from about 2 to 10ish?
This would tend to indicate that you are using different Bayes databases
(running SA as a different user than hMailServer does). To be sure
exactly what is happening, you would need to show us the original
headers vs the headers you get from your manual run.
I admit I don't have any clue about BayesDB so any info will be a huge help.
I will go now and read up.
To summarize, the main point of this thread was about how to run the same
message through with different options. You are telling me getting message
AFTER it goes through hMailServer is ok as it strips the headers. Now I have
to figure out why it scores so low.
Also, I need to do something with BayesDB.
Most of the time, Bayes will take care of itself. Occasionally the
auto-learning routines will get it messed up. You can fix it by running
sa-learn manually to learn batches of spam and ham. In particular, you
should run sa-learn on any message that Bayes scores incorrectly. For
fastest learning, run sa-learn manually on ALL of the ham and spam that
comes through (after manually sorting the messages to make sure they are
all being learned the right way).
--
Bowie
