By learning HAM, am I increasing the probability of HAM tokens?
>>> On 1/13/2006 at 1:17:27 pm, in message
<[EMAIL PROTECTED]>, Bowie Bailey
<[EMAIL PROTECTED]> wrote:
> Clay Davis wrote:
>> If I only ever sa-learn SPAM and never HAM, will the process
>> eventually assign all "tokens" a high probability? Likewise, if I
>> have never sa-learn[ed] any HAM will the first message that is
>> sa-learn[ed] as SPAM assign a 1.000 to all tokens processed?
>
> If you only learn spam, Bayes will never activate. It requires a
> certain number of ham messages and a certain number of spam messages
> before it will start scoring.
>
> Once you have that minimum number, if you continue to train only spam,
> your probabilities will gradually get skewed towards spam and you will
> eventually start seeing false positives. It is normal for bayes to see
> many more spam than ham, so I wouldn't worry too much about keeping it
> balanced, but you should definitely feed it both spam and ham on a
> regular basis.
