Chirs,

Attached, also the before setting it on.

Looks like that the accuracy is less (maybe I interpret it wrong).

On the status page:

Neural Net Accuracy 0.0150, Bitfails 274/1389, Generation time -69

Steffen

From: Support

Sent: Tuesday, May 15, 2012 1:24 AM

To: surgemail-list@netwinsite.com

Subject: [SurgeMail List] Neural Network Beta

If anyone would like to beta test a slight improvement to the spam handling then you may wish to try one of these:

http://netwinsite.com/ftp/misc/v1.zip

http://netwinsite.com/ftp/misc/l1.tar.gz

(binary files, rename your existing surgemail binary first, stop surgemail, replace the binary, and restart)

Then set

g_sf_nnet "true"

then WAIT 30 MINUTES, then type in:

tellmail sf_compare >results.txt

and send me the results.txt file. Also let me know if you the spam detection is better/worse/different over the next

couple of days.

This setting enables a Neural Network that is trained on the sample data surgemail has gathered and figures out how to score messages based on the various features

it detects. In our experiments this seems to give better results than the existing more formal rule based system.

ChrisP.