At 13:23 1/07/2004, Simon Byrnand wrote:
One further thought, since this drastically reduces the number of messages autolearnt, it may take quite some time to reach the 200 ham/spam threshold when first starting out, so if something like this were to go into SA, it might perhaps makes sense for it to be conditional on 200 ham/spam's being learnt.
EG, with less than 200 hams/spams its ok to learn BAYES_99 and BAYES_00 messages, but once 200/200 is reached, it should then start behaving as above...since "dilution" isn't a problem with only a couple of hundred messages learnt, its when you start getting tens of thousands that its a problem...
Bad form to reply to my own message (yet again :) but I realised after clearing my bayes database to give the new system a fresh start, this is already what happens. Until 200 hams and spams are learnt, there is no bayes score to test against, thus the above criteria don't take effect until enough have been learnt for bayes to become active. Perfect :-)
No comments anyone ? nobody willing to try my patch ? :)
Regards, Simon
