Denote the set of features by 
,
and the set of classes by 
.
Given the full training data 
of size 
,
let 
be the number of samples that belong to class 
.
We estimate the frequency of each class by
For a discrete feature 
,
where 
is the smoothing prior
and 
is the total number of
values of .
For a continuous feature ,
the parameters of 
are estimated by maximum likelihood estimation.
For numerical stability, you may instead compute
- distribution: the form of distribution for
- smoothing: additive smoothing parameter
- fit_prior: whether to learn class prior probabilities from data
- class_prior: user-specified prior probabilities of the classes
Check out the documentation listed below to view the attributes that are available in sklearn but not exposed to the user in the software.
- sklearn tutorial on Naïve Bayes.
- sklearn
LogisticRegressiondocumentation.