Migrating from SpamAssassin to Bogofilter

[RW]

On Tue, 04 Jun 2013 23:06:03 -0700
Kip Warner wrote:

> Hey list,
> 
> I'd like to migrate my SpamAssassin setup to Bogofilter, given that
> the former has been no end of headaches for me.

To be fair the problems you reported ended-up being Evolution
plugin bugs.

IMO Bogofilter is a much better candidate for client plugins than
SpamAssassin. If you use one of the normal ways of invoking it SA
usually works without obvious problems, but it's not likely to be
optimal. There are advantages to having it administered by an expert
at an ESP who can devote time every day, and there are advantages
to having a tweaked, well setup local installation, but I think a lot
plugin users are likely to end-up with with the worst of both. 

 
> My current SpamAssassin setup is, or was, made possible by the
> following four user files. They are a white list, two Bayesian
> databases, and some user preferences.
> 
>         ~/.spamassassin/auto-whitelist
>         ~/.spamassassin/bayes_seen
>         ~/.spamassassin/bayes_toks
>         ~/.spamassassin/user_prefs
> 
> Is there a way to train Bogofilter with my SpamAssassin databases?
> I've seen the following on the FAQ,
> 
>   <http://bogofilter.sourceforge.net/faq.shtml#spamassassin>
> 
> , but I believe that's more appropriate for training Bogofilter with
> new / fresh mail, as opposed to what my SpamAssassin has already been
> calibrated with over the years.

If you have the old mail you could just write a script to train
bogofilter from the SA headers, but it's better to train from manually
classified mail anyway - either way I'd suggest stripping X-Spam-
headers. There is some value in training with historic ham, but spam
usually comes along fast enough by itself.

Converting the databases is impractical. Firstly some of the
tokenization is completely incompatible. Secondly SA stores its tokens
as truncated hashes, so you'd need a dictionary which you probably don't
have (AFAIK it's only an option with SQL).