[Spambayes] [ spambayes-Feature Requests-690914 ] Un-classify an email message

[SourceForge.net]

Feature Requests item #690914, was opened at 2003-02-22 08:44
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=498106&aid=690914&group_id=61702

Category: None
Group: None
>Status: Closed
Priority: 5
Submitted By: Carl Nygard (cnygard)
Assigned to: Nobody/Anonymous (nobody)
Summary: Un-classify an email message

Initial Comment:
I'm wondering if the math for adding a message
classified as ham or spam is reversible.  If it is, it
would be a nice feature to reverse the equations and
"subtract" the effect of a certain email.  For example,
if I mis-classify a message, it would be nice to
un-classify it.

I'm guessing that even if the math is close (due to
hysteresis effect??) it might be a useful feature.  But
then, I haven't looked at the code, sorry.


----------------------------------------------------------------------

>Comment By: Mark Hammond (mhammond)
Date: 2003-02-26 23:05

Message:
Logged In: YES 
user_id=14198

This is supported now.  Currently the core does not record
info about how a message has been trained, but assuming the
app knows thus, an unlearn() method exist which does undo
all the maths.

----------------------------------------------------------------------

Comment By: Tim Stone (timstone4)
Date: 2003-02-22 09:39

Message:
Logged In: YES 
user_id=645698

The pop3proxy does not currently support this behavior, though it could 
in theory.  It throws away trained messages at the moment.  This may 
change in the not-so-distant future.

----------------------------------------------------------------------

Comment By: Tim Peters (tim_one)
Date: 2003-02-22 09:28

Message:
Logged In: YES 
user_id=31435

The math is exactly reversible, and the underlying classifier 
has both learn() and unlearn() methods.  Whether you can get 
at them easily depends on the client you're using; for 
example, it's very easy from the project's Outlook client.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=498106&aid=690914&group_id=61702