Tony Meyer wrote: > Is .hammiedb a pickle or bsddb database? It looks like it's a bsddb > database, but you're telling sb_dbexpimp.py that it's a pickle. Try this: > > sb_dbexpimp.py -e -d .hammiedb -f hammiedb.csv
Doh!!! That was it. > It's probably not the corruption of token counts, but of the total messages > trained (at least that was the case ages back when this problem was common). > It shouldn't be possible for training that successfully completes to cause > this problem - if it does, then yes it's a bug (and if you can figure a way > for that to happen, please open a bug tracker on sourceforge and we'll > address it). I restored a .hammiedb from backups from the day before the problem occurred and have continued re-training on errors from there. So far the problem has not re-occurred. I'm wondering if it's a locking issue. The filter is run out of Procmail which uses locking but my train on error script doesn't do any locking by itself. I'm assuming that the Spam Bayes scripts do locking on the hammiedb file to prevent problems. -- Greg Gulik http://www.gulik.org/greg/ greg @ gulik.org _______________________________________________ [email protected] http://mail.python.org/mailman/listinfo/spambayes Check the FAQ before asking: http://spambayes.sf.net/faq.html
