spam cutoff less than neutral?

[Boris 'pi' Piwinger]

Tom Anderson <tanderso at oac-design.com> wrote:

>> >Cutoffs by definition ought to be at or outside of
>> >the min_dev range.  
>> 
>> Not at all.
>> 
>> >Else, min_dev should really be changed to be
>> >consistent with your cutoff philosophy.
>> 
>> It is absolutely consistent. I still don't get you point.
>
>If your min_dev is excluding all tokens between 0.35 and 0.75 as being
>unable to influence a ham/spam decision because they are too
>inconclusive, then it follows that a combined ranking within this range
>is also ambiguous.

Why so? What do you think is the combined ranking? It is
*not* something like a weighted product of the token values.

>If a message classification of 0.55 is definitely
>spam, then an individual token ranking 0.55 should also be indicative of
>spam.  

Not at all. For me a *token* is significat below .2 or above
.8. But a message within that interval is all but unsure. I
have a cutoff of .5 and that works very well.

pi