Am 18.02.2015 um 05:50 schrieb @lbutlr:
On 17 Feb 2015, at 15:46 , Reindl Harald <h.rei...@thelounge.net> wrote:because in a default milter-setup the one and only user is the user which SA and the miler service are running as, hence my script which needs maybe small adjustments for your environment (--no-sync and so on depend on the config, directories needs to exist and permissions for samples needs to be correct)Right. I’m going through your scripts now. They look interesting and with only a few weeks should drop in perfectly.no mysql and what not, just a default bayes-db, two traing-folders for ham and spam and the script for feed new eml-samples as well as the option for ac omplete rebuild based on the current samples, the corpus will stay forever on that machines and samples are named YYYY-mm-dd-number.emlSetting up the spam and ham corpus separately is on thing holding me up right now
how else will you train?? the key part is tell the bayes "this is spam" and "this is ham"
and I’d honestly rather run with a mysql database
should also not be a problem but the "clear" and rebuild could be an issue because in that timeframe the bayes can't be used - the first version did this withou the temp folder ending in warnings and rely on the other SA rules in the meantime
at the begin that was not much a problem with now around 21000 samples this would take 10-15 minuteswell, you don't rebuild regulary but i personally find it an easier way than "forget" to get rid of wrong classified samples or in some cases better not classify some for whatever reason
signature.asc
Description: OpenPGP digital signature
