Next: Learning to Classify Text
Up: Bayesian Learning
Previous: An Example
- We estimated
by
, where 
is the total number of training
instances where 
and 
is the number of these
where 
. - Good estimate except when gets too small (e.g., goes to zero).
- m-estimate of probability:
,
where 
is the prior estimate of the probability and 
is the
equivalent sample size - Can be interpreted as augmenting the actual observations
by an additional virtual samples distributed according to
Patricia Riddle
Fri May 15 13:00:36 NZST 1998