On Fri, 2009-02-27 at 11:45 -0700, Savoy, Jim wrote:
> I would say that about 99% of our training comes from autolearn. I only
> feed (with sa-learn) whatever untagged stuff is gathered up by our control
> group of people and placed in the Public Folders (very little). My original
I suggest not to do train-on-error only. Auto-learn does have a much
higher threshold than required_score, and there are additional
constraints in place, too.
Additionally training low-scoring, verified spam is most likely to help
improve your Bayes results.
> post lamented the fact that the stuff I feed with sa-learn doesn't seem to
> be getting acknowledged. Bayes does say it learned from the messages, and
> the nspam count goes up after I'm done, but the tags afterwards remain the
> same. So it either means that these messages don't have enough discerning
> content, or that I am not using sa-learn properly. I'll try changing my
> username to "exim" the next time I run it. Thanks.
Err, don't try... Just train as the very user SA runs as.
--
char *t="\10pse\0r\0dtu...@ghno\x4e\xc8\x79\xf4\xab\x51\x8a\x10\xf4\xf4\xc4";
main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1:
(c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}
