Re: mailreaver "--undo" ambiguity

ger wrote:
 > Q:
 > On Sat, Jun 13, 2009 at 6:33 AM, Paul Fox<[email protected]> wrote:
 > > hi --
 > >
 > > the CRM...HOWTO.txt file says:
 > >
 > >     --undo     To the extent possible, undo a training with this
 > >                text (cached will be used if possible).  --undo
 > >                requires either --spam or --good as well.
 > >
 > >
 > > it's not clear to me which of --spam or --good i should include
 > > on the commandline.  should it be the way the message _was_ trained,
 > > or the way it _should_ have been trained?
 > 
 > A:
 > Better ever than never? ;-)

thanks -- it is indeed.  :-)

 > 
 > Anyway, it's the '_was_' variant. --undo basically is a <refute>
 > operation and to have it mimic an undo it should be applied to the
 > same db/collection as the previous training-to-be-undone.
 > 
 > Note that you won't get a 'perfect' undo (unless a series of
 > particular conditions are met). Hence, in actual practice, it's an
 > /approximation/ of an undo.

i guess i have a follow-on question:  if the message being "undone"
is in the cache, couldn't information about how it was last trained
be kept in the cache, so that --undo could work without the extra
flag?  i.e., if i train a message as --spam, that fact could be recorded
in the cache.  then when i use --undo, it would effectively apply
the --spam flag for me.

in fact, isn't that information already there?  after training, a message
will be in either "known_good" or "known_spam".  "--undo" can use
that information to train the other way, and to move the message from
one of those subdirectories to the other.  (currently after a
double training the message ends up linked into both known_good
and known_spam.  unless there's a reason for this, it should
always be in just one of those directories.)

note that i'm pretty much ignorant of how crm really works.  i'm just
kibitzing from the sidelines...

paul
=---------------------
 paul fox, [email protected] (arlington, ma, where it's 70.7 degrees)

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