Re: Negative words

[email protected] said:
> - Is there a simple way to trick ifile into having "negative" word
> values (i.e. this "word" should never match folder X ?) 

i.e. if this word appears, then the score for folder X should be 
-infinity?

I think this is possible without changing any code.  What you'd do is give
that word a count of -1 in the database (you'd need to edit your database)
for that class.  ifile smooths counts---it adds one to every count, so if
the database has -1, ifile's smoothed count will be zero. Then, if ifile
sees that word, it will calculate that word's probability for folder X to
be zero.  ifile multiplies all of the probabilities together for each
class.  Thus, folder X will get a zero probability.  The numbers output by
ifile -q are log probabilities. log(0)=-infinity, so if the -1 word
appears, folder X will never be chosen!

There's just one caveat: don't create too many of these -1 count words.  
If you get a e-mail where, for each class/folder, a -1 count word 
appears, all of the classes will get a -infinity score!  Though, I 
guess that's kind of what you'd want to happen anyway, right? :)

Jason