On Tue, Apr 20, 2004 at 11:22:45PM -0700, Daniel Quinlan wrote:
> Duncan Findlay <[EMAIL PROTECTED]> writes:
> 
> > Minor nitpick: I think you mean sets 0 and 2 (i.e. no net with and
> > without bayes)
> 
> mass-check run 1 is for the non-bayes sets.  A single --net mass-check
> is sufficient since you can just strip out the network results and
> generate both of the no-net sets (sets 0 and 1).

Right. I'd forgotten about the need for Bayes to have good scores. (I
did remember that you cant run 2 and 3 with one command)
  
> >> 05/26: generate scores, etc.
> >>
> >> 05/31: announce mass-check run 2 (sets 2 and 3), run until 06/11.
> > 
> > Again: sets 1 and 3 (net with and without bayes)
> 
> mass-check run 2 is for with the bayes sets.  The first set of scores is
> needed to do accurate autolearning for _these_ runs.  It *does* require
> two separate mass-check commands since different sets of scores are
> being used for auto-learning.

Ah... that's where my confusion was. I remembered that we need 3
mass-checks but I couldn't remember exactly which ones. And I agree,
they can be done simultaneously.

> Maybe this should be documented since it seems like we end up
> rediscussing it every release.

We all know we're bad at documenting stuff ;-)

How about we document this in the Wiki perhaps under:

http://wiki.apache.org/spamassassin/DevelopmentStuff

(Any volunteers...?)

-- 
Duncan Findlay

Attachment: signature.asc
Description: Digital signature

Reply via email to