On Wed, June 29, 2011 17:42, Thomas Hruska wrote:
On 6/29/2011 4:08 AM, Erik Ohrnberger wrote:
I gave up on spambayes. It was letting too much spam through. At one time, it was batting 100, but I guess the spammers got smarter or something.
I'm using a gmail account now forwarded to a private (secret) inbox that my outlook pulls down.
Same here (minus the Outlook part). Spammers won because Spambayes didn't keep up, IMO. Letting us define a list of word filters and rules to apply prior to letting Spambayes do its Bayesian thing would be a vast improvement to the product despite what the devs claim. I know my e-mail better than any computer does or likely ever will.
You can do that if you use Spambayes as a procmail filter.
Spambayes also isn't very good about enforcing balanced ham/spam. It works best with equal parts and small numbers of ham and spam but never actually enforces either policy. 99% of what comes into many people's in-boxes is spam. The solution spammers came up with to get through Spambayes was simply to generate more spam and hope people would overtrain Bayesian filters, which is exactly what constantly happens with Spambayes. Bayesian-like filters are hard to work with and really need a trainer analyzer. A good trainer analyzer will keep the filter free of things that would not produce effective results. In this case, the elimination of spam and limited false positives.
amedee@intrepid:~$ ./bin/spamstats Spam: 2974 Ham: 2040 So I have a 3:2 ratio. It could be better, but I'm not complaining.
I can open my mouth here because I've actually done some development on this open source product. Or at least attempted to. My contributions a few years ago were effectively rejected, which pretty much killed any potential future efforts on my behalf. I happen to have some pretty solid C++ COM Outlook add-in development street cred, so the Spambayes devs ruined an opportunity to pick up a Windows developer with the requisite knowledge.
Where is your branch? I assume that you used some kind of public repository for your code? You can always fork...