Training with sa-learn for viruses?

[Nikolaos Milas]

Hello,

I am using amavis with clamav and spamassassin and I am regularly 
training for spam using the command:

    sa-learn --dbpath '/var/amavis/var/.spamassassin' --spam
    /var/amavis/reported-spam

Things work fine with "pure" spam.

My problem is that some highly suspicious email messages that they look 
like they most probably carry infected attachments are not processed as 
spam using the above procedure in order to be blocked in the future.

For example, I am training for 4 messages of which one is true spam and 
the rest are infected mails, but I only get:

    Learned tokens from 1 message(s) (4 message(s) examined)

What should I be doing to train the system and/or its components to 
characterize such mails as "virus" (or at least "spam") and block 
similar ones in the future?

Thanks a lot,
Nick